SuperUROP All SuperUROP Projects ()
spacing
SuperUROP
 
Home
Letters of Recommendation
Login

SuperUROP Projects

"Title: Discrete event simulation of solid-state quantum networks"
Faculty Advisor: Dirk Englund
Mentor(s): none
Contact e-mail: ebersin@mit.edu
Research Area(s): Applied Physics, Communications, Materials, Devices and Photonics, Numerical Methods, Theoretical Computer Science
Over the past decade, advances in quantum hardware have realized the fundamental building blocks of quantum networks, demonstrating high-fidelity single-qubit control and the first multi-qubit links over distances exceeding 1 km [1]. As these technologies continue to mature, there is a growing need to look beyond the single-link level and to consider system-level architectures for building scalable and reliable networks. As these systems are difficult to analyze analytically, one approach is to use discrete event-based simulations [2]. To do so, are working with a platform called NetSquid developed by collaborators at QuTech in Delft. NetSquid is a powerful low-level simulator for quantum networks that follows the paradigm of discrete events. This software can be used to model networks with thousands of end nodes and qubits, as well as repeaters, comparing the performances of different protocols and platforms to allow the design of optimal network architectures for a variety of end applications. In our group, we are using this software to simulate the performance of different entanglement and repeater protocols on a variety of solid-state defect center qubits [3]. Once completed, these simulations will allow for the choice of optimal protocols and qubits for establishing entanglement over a testbed quantum network that connects MIT, Lincoln Laboratory, and Harvard.

The student will be expected to focus on the following tasks:
Set up the Python-based NetSquid software and implement modules for our group’s platform and protocols
Perform simulations to determine optimal parameters for generating entanglement over the Boston-area quantum network
Help theorists and experimentalists implement these results as relevant

Prerequisites: Highly motivated students should have experience programming in Python. A course in quantum mechanics is encouraged, but not required.

References:
[1] B. Hensen et al. Nature 526, 682 (2015)
[2] A. Dahlberg and S. Wehner, arXiv:1712.08032 (2018)
[3] D. Awschalom et al, Nature Photonics, 12, 516 (2018)
"COVID-19: Integrated Predictive Tool for Forecasting Infected Persons and Deaths & Decision Support in Short Term and Future"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Systems, Control and Decision Systems, Numerical Methods, Systems (incl OS, databases, computer security)
The array of predictive models for forecasting numbers of infected persons and deaths by geography and dates are yielding disparate results. Further, few of them provide mechanisms to make estimates based on decisions by various agencies at different points of time. An integrated prediction model and decision support system could provide valuable assistance with respect to current pandemic and similar unexpected situations in the future.

We envision to leverage the MIT experts- Data scientist (predominantly) with guidance of an epidemiologist, economist to build a COVID wave 2 data science predictive model.

We intend to propose the developed predictive model to various US State Governments, agencies for its adoption and operationalization to gain the early warning signs that will help them to proactively manage the crisis situation much better compared to now, that may arise again with a COVID resurgence.
COVID-19: Integrated Predictive Tool for Forecasting Infected Persons and Deaths & Decision Support in Short Term and Future
"Developing Low-cost Tele-ICUs for Addressing Medical Needs of COVID-19 patients"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Communications, Computer Architecture, Computer Networks, Computer Systems, Materials, Devices and Photonics, Programming Languages (incl software eng), Systems (incl OS, databases, computer security)
In order to address several of the diagnostic, treatment, quarantine, and post-treatment medical needs of the COVID-19 patients without infecting medical personnel and other persons, we propose to design, develop, test, and validate an inexpensive, rapidly field- deployable version of the TeleICU that uses mobile phones and portable, simple Bluetooth monitors. this can reduce the cost of a teleICU bed from over $ 30,000 today to under $ 1000 per bed and drastically upscale the operations quickly. This could be combined with novel, low cost ventilator designs created at MIT. The goal is to provide Virtual "Doctors without Borders" capability across organizational and political boundaries.

Link

Link
Developing Low-cost Tele-ICUs for Addressing Medical Needs of COVID-19 patients
"AI-based Approaches for Precision Indoor Agriculture"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computational Biology, Computer Networks, Computer Systems, Machine Learning, Systems (incl OS, databases, computer security)
Due to shrinking farmlands and shortage of resources, various nations are now facing the challenge of feeding their rapidly growing population nutritious and safe food. With their plant-focused light sources and crop-specific lighting recipes, the intent is to enable farmers to adopt Precision Agriculture techniques which allows them to grow crops up to 20x faster and gain up to 20x higher yield in a more sustainable way. Responsive lighting system that adapts its recipe based on plant’s and farmer’s needs will be the next big thing in Agricultural lighting. We propose to conduct research and to design, develop, train, test, and validate new AI-based models and algorithms to predict (a) harvesting time and (b) yield of specific plants in controlled environments such as greenhouses and indoor farms under the influence of artificial lighting. Our approach will be based on multi-model data fusion and machine learning algorithms that employ information from image sensors, environmental sensors, weather stations, historical yields, crop data and other farm data. The main focus will be on tomato and medicinal cannabis indoor crops. AI-based Approaches for Precision Indoor Agriculture
"COVID-19 and Future of Work"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Networks, Human Computer Interaction
COVID-19 has led to greater relevance of working from home and interacting into others who are also working from home, some in other continents. I now schedule meetings from 9 am to 11 pm, Boston time, and had previously coined the term "24-Hour Knowledge Factory" and published papers on it in IEEE, ACM, and other journals; one of them was selected for the Best Paper of 2017 award. COVID-19 and Future of Work
"AI-based Demand Forecasting and Supply Chain Management in COVID-19 era"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Cognitive AI, Computer Systems, Control and Decision Systems, Inference, Machine Learning, Systems (incl OS, databases, computer security)
The coronavirus pandemic has led to an urgent need to analyze data and information collected with automated and semi-automated processes and integrate them with existing computer-based information to create and test forecasting models and predictive methods related to the price, lead time, and stock of different critical items on near-term and continuing basis. AI-based Demand Forecasting and Supply Chain Management in COVID-19 era
"Enhancing Effectiveness and Quality of Healthcare via eICU, AI, and allied mechanisms"
Faculty Advisor: Amar Gupta
Mentor(s): none
Contact e-mail: agupta@mit.edu
Research Area(s): Cognitive AI, Computer Systems, Signals and Systems
This project involves leveraging work of previous students, expanding on it, and doing incremental qualitative analysis and quantitative analysis of data as needed. This super-UROP provides the opportunity to contribute to evolving papers in journal and be co-author of one or more of them. The long-term objective is "Healthcare for All: Better, Quicker, and Less Expensive" and to have a worldwide impact similar to that achieved by earlier members of the group through startups and other channels. Knowledge of qualitative data analysis tools such as NVivo preferred but not essential.

thetech.com/2018/06/07/mit-telemedicine-telehealth-class
"Death by a Thousand Clicks: Where Electronic Health Records Went Wrong: ADDRESSING CHALLENGE OF HEALTHCARE INTEROPERABILITY"
Faculty Advisor: Amar Gupta
Mentor(s): none
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Systems, Human Computer Interaction, Theoretical Computer Science
Link

and

news.mit.edu/2018/removing-health-care-barriers-and-boundaries-amar-gupta-telemedicine-0522

The growing number of medical devices, healthcare algorithms, and the increasing use of telemedicine concepts present a new urgency to the need for surmounting technical and non-technical hurdles that impede the flow of time-sensitive healthcare data and information.
"AI enhanced PRO platform to evaluate and enhance quality of Digital Health approaches and algorithms"
Faculty Advisor: Amar Gupta
Mentor(s): none
Contact e-mail: agupta@mit.edu
Research Area(s): Cognitive AI, Computer Systems, Human Computer Interaction, Materials, Devices and Photonics, Signals and Systems
This project involves designing, developing, and implementing a prototype system that incorporates innovative IT-based approaches for ensuring quality of AI-based and other Digital-based approaches and algorithms for healthcare, possibly by building upon the foundation of the Jeanne Clery Act and incorporating evolving IT approaches for concurrent authentication and anonymity. The theme is to use Patient Based Outcomes (PROs) to look at evolving Digital-Health based approaches and algorithms, especially ones that are not subject to approval by governmental or other authorities. Please take quick look at:
learning-modules.mit.edu/materials/index.html?uuid=/course/6/sp18/6.884#materials
"Intelligent Observing and Multiscale Modeling for Ocean Exploration and Sustainable Utilization"
Faculty Advisor: Pierre Lermusiaux
Mentor(s):
Contact e-mail: pierrel@mit.edu
Research Area(s):
For intelligent ocean exploration and sustainable ocean utilization, the need for smart autonomous underwater vehicles, surface craft, and small aircrafts is rapidly increasing. Applications include scientific studies, solar-wind-wave energy harvesting, transport and distribution of goods, naval operations, security, acoustic surveillance, communication, search and rescue, marine pollution, ocean cleanup, conservation, fisheries, aquaculture, mining, and monitoring and forecasting. Designing optimal paths leads to cost savings, longer operational time, and environmental protection. Our goal is to develop and apply our optimal planning theory and methodology to increase the efficiency of surface craft and underwater vehicles operating in uncertain dynamic ocean conditions. For the first time, we combine environmental forecasting with stochastic control and risk theory, and employ fundamental partial-differential-equations (PDEs) and efficient level-set solutions for exact reachability and path planning. Our novel proposed ocean applications include energy-optimal path planning, optimal environment harvesting, optimal cleanup, and information-optimal exploration and Bayesian machine learning.

The goal of this project is to further develop and apply our exact PDE-based planning theory and data-driven ocean modeling methodology to optimize the efficiency and endurance of ocean vehicles. Possible directions include: i) Implement and apply our theory and schemes for energy-optimal path planning and risk minimization under realistic ocean conditions. ii) Develop and evaluate mission planning for optimal environmental energy harvesting (e.g. solar, wind, and wave energy; algae biofuels) and optimal dynamic ocean cleanup (e.g. marine plastic and litter; oil spills; natural and man-made sediment plumes); or iii) Develop information-optimal theory for efficient scientific exploration and Bayesian machine learning of ocean model parameterizations and turbulence closures.
Intelligent Observing and Multiscale Modeling for Ocean Exploration and Sustainable Utilization
"Brain Tumor Modeling and Learning"
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
Mathematical models, typically a system of ordinary and/or partial differential equations, can provide considerable insight into the dynamics of biological systems. We attempt to model the evolution of glioblastoma multiforme (GBM), the most common (and most aggressive) type of human brain cancer, but significant amount of uncertainty exists in the functional form of model equations and parameterizations. This is due to the complexity and lack of understanding of the processes involved, along with patient-specific differences in brain cell density and geometry. These challenges motivate the objective of the present work in which we try to integrate noisy and sparse (in time) data extracted from magnetic resonance (MR) images of the patient into the model using a dynamics-based Bayesian learning framework. Accurate predictive modeling could be clinically helpful in many ways, for example in important decisions related to scheduling surgery and preparing a more patient-specific therapy plan.

The SuperUROP project would be to implement the mathematical model for GBM in our rigorous PDE-based machine learning framework, perform image processing on the magnetic resonance (MR) images to extract data, and visualization of results. The project involves new avenues of collaboration between doctors, biologists, and predictive modelers. The student will gain first-hand experience in powerful numerical analysis and data assimilation techniques, which are applicable to a variety of research and societal needs, including applications in energy, food, sustainability, and security.
Brain Tumor Modeling and Learning
"Machine Learning of Uncertain Dynamical Models"
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
Our understanding of many physical systems in the world is imperfect. This is why it is important to account for the uncertainty when making predictions of the weather, ocean currents, river pollution, or complex fluid flows, to name a few. In this particular project, the dynamical equations governing the physical system are either not well known or unknown, but some data time series are available. To recover or discover the dynamical equations, Bayesian methods combining model prediction and data assimilation may be used. Another approach uses deep neural networks to make predictions of the system evolution without learning, or while learning, the model equations.

This SuperUROP will work on the various methods developed by our group to learn the governing model equations and functional formulations from sparse data in a Bayesian sense. We have also been utilizing more traditional machine learning techniques such as Gaussian processes and neural networks to learn the model dynamics through data driven observations / sampling. The SuperUROP will work with our team to further develop and apply Bayesian and deep learning methods.
Machine Learning of Uncertain Dynamical Models
"Interdisciplinary Data Assimilation and Machine Learning"
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
Ocean observations are limited. Extracting all of the pertinent information from observations is thus critical for many interdisciplinary ocean applications. This is especially true for coupled physical-acoustical, physical-biogeochemical, and physical-sea-ice ocean dynamical systems. Applications abound from multi-resolution underwater GPS and ocean bathymetry estimation to machine learning of sea-ice dynamics and predicting the health of marine ecosystems such as coral reefs.

This SuperUROP will work with our group on the nonlinear non-Gaussian estimation of the state, parameters, and parameterizations of such interdisciplinary systems using our modeling systems and Bayesian filtering and smoothing methods. The concepts of mutual information, predictability, and optimal learning are likely to be used. Examples will highlight how, when, and why it is useful to exploit the governing nonlinear dynamics and to capture the non-Gaussian structures in such interdisciplinary studies.
Interdisciplinary Data Assimilation and Machine Learning
"Hybrid Discontinuous Galerkin (HDG) Finite Element Schemes"
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
We are developing high-order hybrid discontinuous Galerkin (HDG) finite element methods for high-fidelity fluid and ocean predictions. We plan to further develop, utilize, and refine these codes to accurately model multi-scale and non-hydrostatic ocean phenomena. Additionally, high-order HDG modeling of real and idealized flow around vehicles can be used to better estimate the “power expenditure versus velocity” laws of underwater vehicles. Lastly, the HDG methodology is well-suited to unstructured problem geometries and to both fluid and solid mechanics. We plan to use the code to investigate the physics and modeling of sea-ice for realistic ocean flows.

The SuperUROP will work on development of the group’s HDG finite element code in Python and C++, and the extension of the code to new applications, such as those mentioned above. Further work includes development and improvement of HDG-specific parallel computing algorithms aimed at efficiently solving large time-dependent and 3D-in-space problems.
Hybrid Discontinuous Galerkin (HDG) Finite Element Schemes
"Mobile Applications for Ocean Predictions and Marine Autonomy."
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
Many of us check the weather on our smartphones (or ask Alexa) to decide what to wear today or what to do the next weekend. Smart devices are becoming ubiquitous and they track many human activities to help us make better decisions. However, out in the deep seas, we don’t yet have the luxury of having ocean information at our fingertips. Fortunately, this situation is changing and we at the MSEAS group have recently developed capabilities for smart autonomous monitoring and modeling of our oceans. Specifically, we have developed a suite of science- and engineering-based data-driven stochastic prediction systems to forecast ocean conditions, optimal ship routes, probable fishing zones, and pollution.

We now seek a SuperUROP to develop web and mobile applications that scale and deploy our modeling results to our end-users. The proposed work involves developing front-end (web and mobile interface) and back-end (distributed and cloud) technologies. For the front-end, we plan to develop an iOS, android, and web apps. A mock-up of our proposed web front-end is shown in Fig. 1. For the back-end, we will employ xarray, dask and kubernetes to process our modeling results and serve it to the front-end. Candidates are expected to possess excellent programming skills (e.g. python), and an interest in working with big data, big compute and AI for environmental conservation. Experience with creating mock-ups or user-interface design is another advantage.
Mobile Applications for Ocean Predictions and Marine Autonomy.
"Energy-Optimal Path Planning for Vehicles Operating in Uncertain Environments"
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
The use of autonomous vehicles is growing in a wide range of applications such as distribution of goods, security, acoustic surveillance, search and rescue, oil and gas exploration, land and ocean floor mapping, conservation, and ocean monitoring and forecasting. In marine operations, due to uncertainties, ocean sensing is often at the heart of most autonomous missions. For missions that involve both sensing and exploration, long endurance and low energy cost are crucial requirements. This project is concerned with fundamental research towards quantitative energy-optimal planning for ocean and air vehicles operating in uncertain dynamic environments.

Specific SuperUROP objectives can be to: (i) apply our stochastic PDE-based optimization for realistic missions using 3D-in-space and time-dependent ocean simulations; (ii) contribute to the theory and schemes for energy-optimal path planning under uncertain flow conditions; (iii) apply mission planning schemes for optimal environmental energy harvesting; (iv) employ high-fidelity models of flows around vehicles to improve energy expenditure estimates; (v) include our work on pattern coordination, dynamic obstacle avoidance, and evader-pursuer problems as part of the energy-optimal path planning. The focus is on the ocean domain but our objectives are relevant in other domains, e.g. efficient drone operations, optimal storm avoidance, and landing in airports.
Energy-Optimal Path Planning for Vehicles Operating in Uncertain Environments
"Advanced Lagrangian Predictions for Hazards Assessments (ALPHA)."
Faculty Advisor: Pierre Lermusiaux
Mentor(s): none
Contact e-mail: pierrel@mit.edu
Research Area(s):
Hazards due to the fundamental process of advection of natural and anthropogenic material in environmental flows are ubiquitous and profoundly impact society; preparedness and effective response can save many lives, untold environmental damage and enormous financial cost. On a day-to-day operational level, search- and-rescue operations at sea rely critically on correctly modeling and interpreting flow transport in order to inform life-or-death decisions. Understanding how flow transport is organized and predicting where things go in complex environmental flows remains a formidable scientific challenge, however, due to unsteady nonlinear and multiscale flows, ambiguities in defining material transport, multiple sources of uncertainty, the difficulty of identifying and acquiring pertinent data to assimilate into models, the variability of predictions across different models, and the complexity of analyzing vast data sets and visually representing the results.

This SuperUROP project would be to contribute to a research team effort that addresses these science issues by exploiting and advancing recent fundamental breakthroughs in four-dimensional (3D+time) Lagrangian methods. The developed approaches to study Lagrangian transport are employed in ideal as well as realistic applications. These methods are customized for the specific type of hazards to be modeled (such as search and rescue, sediment plumes, debris, etc.). Finally, these cutting-edge Lagrangian methods are coupled with data-driven modeling to learn the governing equations, quantify and predict the key transport processes and structures during regional flow-based hazards in the ocean and atmosphere.
Advanced Lagrangian Predictions for Hazards Assessments (ALPHA).
"Reading to Write: Using NLP to enhance online annotation and collaborative argument development"
Faculty Advisor: Kurt Fendt
Mentor(s):
Contact e-mail: fendt@mit.edu
Research Area(s):
Online annotation tools such as the MIT-developed Annotation Studio (annotationstudio.org) have been used successfully to support college textual annotation and collaborative reading. On the other hand, these tools have the potential for creating a “writing space” in which student annotations from one or more texts may be organized and used as the basis for the development of arguments by individuals or by collaborating groups.
Our goal is to research and implement a variety of NLP approaches to a) give students visual insights into their writing; and b) support writing instructors in reviewing student essays. To this end, we will expand Annotation Studio and the prototype of Idea Space into a versatile online tool that will seamlessly integrate the processes of close reading, annotation, writing, and reviewing.

SuperUROPs will work closely with the PI, scholars in the Writing, Rhetoric, and Professional Communication (WRAP) program, and CSAIL faculty to research, evaluate, and implement NLP techniques into Annotation Studio and test it with students and faculty in SHASS.
Reading to Write: Using NLP to enhance online annotation and collaborative argument development
"User-centric Healthcare Platform"
Faculty Advisor: Lalana Kagal
Mentor(s):
Contact e-mail: lkagal@csail.mit.edu
Research Area(s):
Applications that track, and analyze health and genetic data such FitBit, RunLogger and 23andme are becoming commonplace. These applications store the data they collect in centralized repositories that can be thought of as "data silos". Similarly, healthcare providers use a range of different applications to make electronic health records (EHR) available to users while storing them centrally. Each application controls its own data and often has its own authentication and policy mechanisms. As a result, users cannot easily switch between similar applications that would allow interesting analysis or integration of their health data, or switch from one data storage service to a different one. It is also difficult for a user to combine data across these silos to identify interesting correlations such as those between sleep patterns and diet or monitor day-to-day health statistics.

We are developing a new Web architecture, Solid, in which users have control of their data and the Web is user and data centric. Solid is an open-source project, principally based on well-established standards, and orientated in such a way that data can be managed by the originator and shared appropriately. We are interested is investigating how Solid can be used as a decentralized infrastructure for health data storage that will empower individuals by giving them control over their health data and allow privacy-aware research and analysis for personal and societal benefits.

Experience with JavaScript development is a requirement, and familiarity with Linked Data is a plus. Please send your CV along with a description of relevant research projects to <lkagal@csail.mit.edu>.
User-centric Healthcare Platform
"PrivacyML - Privacy Preserving Frameworks for Machine Learning"
Faculty Advisor: Lalana Kagal
Mentor(s): Vaik Mugunthan
Contact e-mail: lkagal@csail.mit.edu
Research Area(s): Computer Systems, Human Computer Interaction, Machine Learning, Systems (incl OS, databases, computer security)
Machine learning plays a major role in many of the services and products we use today such as text and voice assistants, social media, location services, advertisements, and even auto-complete. It is used to deduce novel information and patterns from existing data. With the advent of novel machine learning optimizations and deep learning, all kinds of analysis can be performed on data that may compromise the privacy of the individuals in the dataset. Though anonymization is the first line of defense, re-identification and reconstruction based attacks are
becoming more prominent due to the vast amount of data available. At the same time, with deep neural nets, models and systems are able to memorize information from data sets that can be later extracted through GANs and cycle GANs. Though machine learning increases the utility of data, it has an adverse effect on privacy.

One approach to preventing these kinds of privacy violations is to develop data and domain specific privacy preserving versions of machine learning algorithms. However, this requires data scientists and analysts to be intimately familiar with privacy techniques, be constantly aware of it, and understand how to customize it for different domains and applications. We propose an alternate approach that of generalized frameworks that enforce privacy internally and enable different kinds of machine learning algorithms to be used that are automatically privacy preserving. We believe that this decoupling of privacy and machine learning is important as it gives analysts the freedom to analyze data for various purposes without worrying about violating privacy.

Within this area, we have several projects - privacy preserving federated learning (when distributed parties want to collaboratively develop a model), privacy while analyzing unstructured data (such as doctors notes, insurance claims, and blogs) and media, and privacy when learning over knowledge graphs.

Familiarity with machine learning is required and some experience with security or privacy technologies is a plus. Please send your CV along with a description of relevant research projects to <lkagal@csail.mit.edu>.
PrivacyML - Privacy Preserving Frameworks for Machine Learning
"Using Blockchain Smart Contracts for Automating Legal Agreements"
Faculty Advisor: Lalana Kagal
Mentor(s): none
Contact e-mail: lkagal@csail.mit.edu
Research Area(s): Cognitive AI, Computer Systems, Human Computer Interaction, Inference
Formal legal contracts govern many real-world applications, ranging from data sharing systems to complex financial transactions. For all types of agreements, lawyers employ similar, manual drafting and revision processes. When disputes arise, lawyers must re-read agreements to present arguments for settlement or trial. Each of these manual steps during creation, validation, and enforcement of an agreement incurs significant legal fees for lawyers' time. Moreover, no manual check will catch all errors, and inadequate agreements resulted in multi-million dollar lawsuits.

While some have attempted to use natural language processing to automatically interpret legal agreements, such systems are limited to certain legal domains and are not focused on detecting errors or easing with litigation. This limitation prevents automation of legal processes, and by extension, significant reduction of legal expenses.

Blockchain technology is proving to be very useful in this domain specifically the use of smart contracts, which are pieces of code stored on a blockchain that are triggered by blockchain transactions and that read and write data on that blockchain. By representing agreements and contracts as smart contracts on the blockchain, the vast paperwork and oversight needed to maintain these contracts will be eliminated leading to reduced risks, lower administration and service costs, and more efficient business processes. The problem that remains is how to capture the semantics of legal contracts and enable their automated verification and adjudication. This project deals with this gap by working on high level constructs for specifying legal contracts to enable them to be translated to verifiable smart contracts that can be adjucated on the blockchain.

Familiarity with Ethereum is a plus but not required. Please send your CV along with a description of relevant research projects to <lkagal@csail.mit.edu>
Using Blockchain Smart Contracts for Automating Legal Agreements
"Assessing the impact of significant hurricane events on the sustainability of energy generation infrastructure"
Faculty Advisor: Saurabh Amin
Mentor(s):
Contact e-mail: amins@mit.edu
Research Area(s):
Natural disasters such as hurricane strikes, cyclone bombs and floods are occurring more frequently due to the climate change. These disasters can often damage bulk power generators and cause regional power outages, which can last temporarily or for a prolonged time period. In this project, our goal is to assess the impact of recent hurricane events on the survivability and composition of bulk power generation sources. We also seek to understand the strategic choices made by generation owners in re-building their infrastructure, and the potential to diversify into renewable energy sources. The project involves collecting power generation data during and after recent major hurricane events in the US. In addition, hurricane track ensembles and wind field simulations will be used to estimate and validate damage to generation infrastructure.

The overall objective is to assess severity of damage using metrics such as the maximum loss of generator capacities, the rate of recovery after the hurricanes, and the production mix between conventional and renewable energy sources in regions that are prone to hurricane-induced damage. The project involves developing a statistical model to analyze the causal relationship between the damage and the significant features of the hurricane such as the wind speed, landing location, affected area, etc. We also study whether or not the level of damage and recovery rate depends on the composition of energy sources. This analysis will help us make better decisions for technology investment and infrastructure maintenance in order to improve the resiliency of power systems.

Prerequisite knowledge: Ability to code in Python or R, Basic knowledge of statistics and probability. Background in machine learning and causal inference is a plus.
Assessing the impact of significant hurricane events on the sustainability of energy generation infrastructure
"Assessing the Demand for Urban Air Mobility through agent-based simulations of prototype cities"
Faculty Advisor: Moshe Ben-Akiva
Mentor(s): Md Sami Hasnine
Contact e-mail: mba@mit.edu & hasnine@mit.edu
Research Area(s):
The aim of the proposed research project is to investigate the impacts of urban air mobility (UAM) (e.g., flying taxi) services on transportation in the context of North American cities, and to quantify UAM demand and in particular, its relationship with service pricing, operational configurations and other city-specific characteristics. This will be achieved through large scale agent-based simulations of selected North American prototype cities of interest. The prototype city models will be constructed for simulation in SimMobility, and will include the modeling of UAM services on the demand and supply sides. The analysis of the pricing strategies will provide us insight into the potential market penetration of UAM services, and also yield insights into the location of vertiports based on the spatiotemporal distribution of demand. The travel range will depend on the geographical area of the selected cities, which constitutes intra-urban travel.

Prerequisite Knowledge: Experience in programming, preferably experience in C++, SQL, python is a plus.
Assessing the Demand for Urban Air Mobility through agent-based simulations of prototype cities
"Behavioral response to new Shared Autonomous Vehicle and mobility services offered in a Mobility-as-a-Service platform across different types of cities"
Faculty Advisor: Moshe Ben-Akiva
Mentor(s): Jinping (Jenna) Guan
Contact e-mail: mba@mit.edu & jinpingg@mit.edu
Research Area(s):
The project will 1) Assess behavioral changes in response to the introduction of new shared autonomous vehicle (SAV) and mobility services, including modalshifts, time-of-day patterns, and trip purposes, across different types of cities; 2) Quantify the system impacts of new autonomous vehicle (AV)/mobility services, including congestion and vehicle miles travelled, across different types of cities; 3) Find optimal solutions for AV/mobility deployments balancing the interests of customers, fleet operators, and cities that could be scaled across cities of the same type.

Prerequisite Knowledge: knowledge about SimMobility, Shared Autonomous Vehicle and Mobility-as-a-Service (simply search online and learn by yourself); experience in programming, preferably experience in C++, SQL, python.
Behavioral response to new Shared Autonomous Vehicle and mobility services offered in a Mobility-as-a-Service platform across different types of cities
"Experimentation on Hydraulic Fracturing and Facture Flow"
Faculty Advisor: Herbert Einstein
Mentor(s):
Contact e-mail: einstein@mit.edu
Research Area(s):
Hydraulic fracturing and fracture flow are governing processes in Hydrocarbon (oil and gas) extraction. They are equally if not more important in EGS (Engineered Geothermal Systems) with which sustainable energy can be produced. The laboratory project involves both, but is particularly relevant regarding EGS, where not much preceding research exists.

Very innovative equipment exists in our labs that has been and will be used for this research:
- Hydraulic fracturing with external load application and pressurization of fractures with different fluids. Visual observations with high-speed camera and acoustic emission sensors.
- Fracture flow experiments with Hele Shaw cells again allowing one to apply external loads and internal pressures as well as using visual observations.

The unique characteristic of both types of experiments is the combination of visual and indirect observation. This provides the basis for models that can be used in the field where only indirect observations are possible. The project not only involves running laboratory tests but analysis of the results using principles of mechanics and statistics.

It is very important that the UROP student will work closely with graduate research assistants and Professor Einstein who have and are extensively working with undergraduate students on research projects.
Experimentation on Hydraulic Fracturing and Facture Flow
"Ultra-Small-Scale THz Circuits"
Faculty Advisor: Ruonan Han
Mentor(s):
Contact e-mail: ruonan@mit.edu
Research Area(s): Circuits
Integrated circuits operating at the terahertz frequency will serve as the infrastructure for the next generation radar imaging and high-speed communications. It is noteworthy that these applications rely on an assumption that, as the wavelength decreases, a larger number of coherent radiator elements will be implemented inside the same physical size. Such a configuration will then generate narrower radiation beamwidth (i.e. pencil beam), which increases the imaging resolution and communication signal-to-noise ratio.

That assumption is unfortunately not fully valid. The down-scaling of the RF-frontend circuits fails to nicely follow the increasing operation frequency. And normally, the size of the circuits is much larger than a half-wavelength by half-wavelength space. That prevents a dense array placement of these circuits, and the related system (e.g. phased array) will generate a beam with undesired sidelobes.

The fundamental reason for the above problem lies within a change of design methodology, from lumped-element (e.g. capacitors and inductors) approach to distributed-element (e.g. transmission lines) approach, when the circuit's frequency approached tens of gigahertz. That was because lumped elements were considered to have poor scalability and efficiency. In this project, the SuperUROP will be responsible of revisiting these discussions, and exploring new design approaches to re-use the small size lumped elements in THz integrated circuits. A Ph.D. student from Prof. Han's group will be assigned as the mentor of the SuperUROP.
Ultra-Small-Scale THz Circuits
"Planning dashboard for integrating autonomy into urban systems"
Faculty Advisor: Cathy Wu
Mentor(s):
Contact e-mail: cathywu@mit.edu
Research Area(s):
The aim of this project is to produce a planning dashboard suitable for decision makers in the urban system to make informed choices about the integration of autonomy into their cities. Specifically, through a partnership with Taichung City, Taiwan and the National Center for High-Performance Computing in Taiwan, we aim to scope out the decision space for integrating automated traffic signals and autonomous vehicles into a dense urban area, such as the fraction of lights or vehicles needed for effective traffic management, the degree of needed compliance or adoption, and their corresponding energy consumption change, cost and other impacts. This project would involve a combination of literature review, engaging with practitioners from the public and private sectors, wire-framing, implementation of a dashboard, and integrating it with an underlying reinforcement-learning-based decision system. Depending on student interest, there are also opportunities to participate in important underlying modeling questions, such as developing machine learning pipelines for modeling of regional traffic scenarios and driving behavior.

Requested prerequisite knowledge: visual design experience, excellent communication skills, web programming experience. Interest in transportation systems or urban planning is expected. Exposure to machine learning and control systems is a plus.
Planning dashboard for integrating autonomy into urban systems
"Predation and spatial structure in the healthy skin microbiome"
Faculty Advisor: Tami Lieberman
Mentor(s):
Contact e-mail: tami@mit.edu
Research Area(s):
The Lieberman Lab is looking for a SuperUROP to join us to explore microbial-phage coevolution in the healthy human skin microbiome. The Lieberman Lab has been studying microbial evolution and ecology on sebaceous skin (e.g. face), a diverse ecosystem of bacterial, fungal, and viral inhabitants. We have discovered that skin pores operate as distinct ecological units and promote intraspecies diversity. This project will use whole-genome sequencing and culture-based approaches to explore how phage dynamics and evolution drive bacterial community colonization and adaptation within a single pore.

Requested prerequisite knowledge: Student should have a basic understanding of cellular biology
Predation and spatial structure in the healthy skin microbiome
"Hydrothermal processing (HTP) and computational modeling for turning waste into useful materials"
Faculty Advisor: Markus J. Buehler
Mentor(s):
Contact e-mail: mbuehler@mit.edu
Research Area(s):
Black liquor, an aqueous by-product in the paper industry with high organic content (primarily lignin and carbohydrates), has been studied for better utilization for a long time, with few solutions found thus far. In the paper industry, while fibers in wood are separated out for papermaking, while lignin is a leftover component, which is difficult to deal with, and often either burned or disposed. The hypothesis of this work is that we can use the rich chemical composition of these waste streams and transform them into novel engineered materials, offering new pathways to create useful forms of matter.

Hydrothermal processing (HTP), which typically only uses water, heat and pressure, has been recognized as an environmentally friendly method for synthesis and processing materials. Preliminary work in my lab has already demonstrated that it can be a powerful approach, especially when combined with molecular modeling, to design better materials from less. Indeed, computational modeling can be used not only as a supplementary tool to experimental work, but also a powerful tool for bottom-up design, which can lead to de novo approaches and solutions.

In this project, we will apply both experimental and modeling methods to explore possible solutions for black liquor. HTP as well as other experimental methods may be applied to convert black liquor into valuable products, such as bio-based oil that can be used directly or upgraded into biofuels, or other more complex chemicals. Molecule dynamics (MD) simulations and density functional theory (DFT) will be conducted to 1) explain the experimental data; and 2) investigate possible design solutions from the bottom up, thus leading to de novo solutions to turn waste into useful, functional materials. In this project, we hope to demonstrate the synergy of experimental and modeling methods, and also find de novo solutions for industrial problems, to address environmental waste.

Requested prerequisite knowledge: Strong interest in hands-on experimental work and a passion for environmental science and engineering at the intersection with materials engineering; some experience in a chemistry lab preferred; basic organic chemistry/chemistry knowledge. Students will be fully trained, but some knowledge/concepts of molecular dynamics (MD) simulations and/or density functional theory (DFT) is advantageous.
Hydrothermal processing (HTP) and computational modeling for turning waste into useful materials
"Intragastric satiety generating robot"
Faculty Advisor: G. Traverso
Mentor(s): Neil Jia
Contact e-mail: zixunjia@mit.edu
Research Area(s):
Obesity lays heavy burden on the health sector of the U.S. with 31.8% of the population is overweight. Treatments for obesity include behavioral interventions, diet and exercise, medications, endoscopic and surgical interventions. However, current treatments either lack long-term efficiency or have recognized severe complications. This project exploits gastric mechanical signaling of satiety neuron-hormone pathways, and aims to develop applied interventions for the modulation of satiety to relieve the burden of obesity. The project explores synergistics approaches which not only stimulate local mucosa in the gastric cavity but also stimulate the neuronal pathways. This work involves significant prototyping, ex vivo evaluation and in vivo long term testing.
This Super UROP project aims to design, fabricate and test the robot in vivo in a swine model. The Super UROP student is expected to help design/fabricate part of robot, program relevant code and participate in the in vivo/ex vivo experiments upon progress. Project requires background on robot design, fundamental control theory and programming, no medical/biological research experience is expected.
"Learning to Evaluate Product Creativity"
Faculty Advisor: Faez Ahmed
Mentor(s):
Contact e-mail: faez@mit.edu
Research Area(s):
The creativity of engineering products is often assessed by human experts. However, experts may have limited bandwidth in evaluating many items and often have a low inter-rater agreement. In this project, we want to learn mathematical functions using interpretable machine learning methods, which can not only augment creativity evaluation by human experts but also decode the factors which help explain human decisions. Your job: You will work closely with the professor and learn to use cutting edge machine learning methods for design representation and supervised learning. We are looking for a student who is comfortable in programming using Python or MATLAB and is interested in machine learning applications for design. Learning to Evaluate Product Creativity
"Multiresolution simulations and modeling of 3D turbulent fluid flows"
Faculty Advisor: Wim M. van Rees
Mentor(s): Thomas Gillis
Contact e-mail: wvanrees@mit.edu
Research Area(s):
A hallmark of complex 3D turbulent fluid flows is the non-linear interaction between structures at different scales, ranging from inertial scales of large vortical eddy’s to the viscous scales where energy dissipates. In our group, we are developing a new multiresolution computational framework that combines wavelet-based multiresolution analysis to simulate 3D flows around immersed obstacles with unprecedented complexity.
In this superUROP project, we are looking for a computationally-minded student to enhance this multiresolution framework with turbulence modeling techniques, in order to extend the range of Reynolds numbers that can be simulated. In particular, we aim to incorporate uRANS and LES models, which are widely used techniques in many aero- and hydrodynamic engineering designs. The project will start by researching and implementing some existing approaches for uniform resolution grids, and then working towards models that fully exploit the multiresolution framework, without losing consistency. We expect the ideal candidate to have a working knowledge of C and/or C++, be familiar with basic computational engineering principles, and have an interest in scientific computing, parallel programming, and computational fluid dynamics.
Multiresolution simulations and modeling of 3D turbulent fluid flows
"Competing (mis)information narratives and human behavior in pandemics"
Faculty Advisor: Una-May O'Reilly
Mentor(s): Erik Hemberg
Contact e-mail: alfa-apply@csail.mit.edu
Research Area(s):
We propose the use of agent-based modeling to determine the effects of
COVID related health information to the public. The current global
health recommendation for mitigating the COVID spread is social
distancing. The impact of this method is based on how citizens
interpret, act and comply on health information. There is a disconnect
in both the messaging and obedience to public directives, which can
have massive implications. We investigate a data-driven, agent-based
modeling method based on human moral cognition for effective public
health communication. We will construct of models from regions that
are ahead in the COVID spread to provide realistic decisions for other
regions. This can cover the different stages of a viral pandemic, from
the outbreak, to apex, and the return to pre-pandemic levels of
society without exhausting existing resources.

Please include in your email: the title of the project you are interested in, your CV, graduation year, relevant courses and grades, relevant research or experience
Competing (mis)information narratives and human behavior in pandemics
"MOOCs Reflecting in online learning"
Faculty Advisor: Una-May O'Reilly
Mentor(s): Erik Hemberg
Contact e-mail: alfa-apply@csail.mit.edu
Research Area(s):
Does learning online provide you with the same opportunity for
reflection as in a normal classroom? Online learning has many
benefits, but there is still room for improvement. We research the
role of active reflection on the learning for online learners and
how that impacts learning outcomes. This requires data science and
natural language processing to handle the large volume of free text
answers that students provide.

Desired NLP course

Please include in your email: the title of the project you are interested in, your CV, graduation year, relevant courses and grades, relevant research or experience
MOOCs Reflecting in online learning
"IDAS Synthesizing programs with evolution and formal methods"
Faculty Advisor: Una-May O'Reilly
Mentor(s): Erik Hemberg
Contact e-mail: alfa-apply@csail.mit.edu
Research Area(s):
How can better software be automatically engineered? Is it possible to tell the
computer what to implement a task instead of how to implement it?
To proceed in this direction, we will combine an AI-based approach
based on stochastic search heuristics with formal program
synthesis methods. Our goal is to generate correct software based on a "what is needed"
specification.

Requires a software engineering course.

Please include in your email: the title of the project you are interested in, your CV, graduation year, relevant courses and grades, relevant research or experience
IDAS Synthesizing programs with evolution and formal methods
"CHASE Cyber-hunting with public knowledge and AI"
Faculty Advisor: Una-May O'Reilly
Mentor(s): Erik Hemberg
Contact e-mail: alfa-apply@csail.mit.edu
Research Area(s):
Do you wait for your adversary, or do you go and find him? Today,
cyber defenders need to be active when protecting their cyber
assets. Actively hunting for adversaries is one way of preventing
attacks. But how do you find the adversary? There are currently public
resources describing cyber tactics, techniques and procedures and by
utilizing this public threat data with Artificial Intelligence you
will research how to improve cyber defenses.


Please include in your email: the title of the project you are interested in, your CV, graduation year, relevant courses and grades, relevant research or experience
CHASE Cyber-hunting with public knowledge and AI
"Statistical Models of Computer Programs"
Faculty Advisor: Una-May O'Reilly
Mentor(s): Shashank Srikant
Contact e-mail: alfa-apply@csail.mit.edu
Research Area(s):
We aim to understand and analyze programs from a data-driven perspective.
Imagine solving hard program analysis tasks like finding bugs and vulnerabilities in them. We view this as a problem in training machine learning models to capture structured information (loops, data and control dependencies, parse trees of programs) to predict properties like bugs and vulnerabilities.
This is a rich and nascent problem space, and we are actively working on several aspects, including representation learning, evaluating different architecture designs, and on even synthesizing programs from generative models.

Pre-requisites: A mix of systems and ML inclination. Preferable if the applicant has taken one or multiple of these --
Systems - 6.009, 6.031, 6.033, 6.035, 6.820, 6.858
ML/NLP - 6.008, 6.034, 6.036, 6.806/6.864, 6.867,



Please include in your email: the title of the project you are interested in, your CV, graduation year, relevant courses and grades, relevant research or experience
Statistical Models of Computer Programs
"Machine Vision System for Robotic Endovascular Neurosurgery"
Faculty Advisor: Xuanhe Zhao
Mentor(s): Yoonho Kim
Contact e-mail: zhaox@mit.edu, yoonho@mit.edu
Research Area(s):
Based on our recent development of magnetically steerable guidewires -- thread-like soft continuum robot s(see the attached image) -- we are currently developing a teleoperated robotic platform for endovascular neurosurgery (e.g. stroke or aneurysm treatment in the brain’s blood vessels). To basic idea is to use a robotic manipulator to magnetically steer our guidewire upon remote control to enable robotic endovascular neurosurgery. In line with this overarching goal, our SuperUROP project is focused on developing an imaging system which will be coordinated with a robotic actuation and control system for visual feedback to monitor and track the position and orientation of the guidewire in the vascular phantom in real time during the robotic manipulation. More specifically, we will be investigating relevant topics such as 3D Image Reconstruction and Digitally Reconstructed Radiography to simulate actual clinical settings in which real-time fluoroscopy based on X-ray imaging is used. The developed imaging system will be used for learning-based control and motion planning of our robotic arm. We are envisioning working closely with medical companies and clinicians over the course of project. Students who have experiences or interests in robotics, machine vision/learning, medical imaging with basic knowledge of programming (e.g. Python) are encouraged to apply. Machine Vision System for Robotic Endovascular Neurosurgery
"Design, construction and testing of a feedback controlled, eccentric mass shaker"
Faculty Advisor: Kim Vandiver
Mentor(s):
Contact e-mail: kimv@mit.edu
Research Area(s):
This project will involve designing and building dual shaft, counter-rotating, eccentric mass shakers. The counter rotating shafts make it possible to have two rotating eccentric masses, which produce a periodic force in only one direction. A key element of a precision controlled shaker is good speed control. A motor controller with feedback is an essential part of the system, to prevent the shaker from excessive rotation rate fluctuation, caused by periodic external torque.

The shaker has two immediate applications:
1. 2.003 vibration demonstrations. I made a 6-degree of freedom, vibration demo for 2.003 last year. Last fall it was used for recitation demos with a crude rotating mass shaker. I would love to be able to do really precise in-class demos.
2. Measurement of damping of large flexible electrical cables. My research on flow-induced vibration of cables and flexible production risers includes measurement of damping of multiple layered armored cables. The results are needed for the design of cables and risers in offshore energy projects. We need to design a shaker that can be used to precisely test large samples--50 m long and 0.1 to 0.2 m in diameter.

I believe this would be a good project for a mechanical engineering student who likes to design and build.

Prospective candidates should contact Prof. J. Kim Vandiver, kimv@mit.edu
Design, construction and testing of a feedback controlled, eccentric mass shaker
"Studying the Role of Dynamics in Electrolytes for Lithium Ion Batteries"
Faculty Advisor: Yang Shao-Horn
Mentor(s): Jeffrey Lopez
Contact e-mail: shaohorn@mit.edu, jlopez1@mit.edu
Research Area(s):
New electrolytes offer immense opportunities to transform electrochemical
energy storage technologies including batteries and fuel cells by
increasing safety and reducing costs. However, fundamental gaps exist in
our understanding of the molecular origins of ion mobility in liquid,
polymer, and inorganic media. On this project you will have the opportunity
to join a multidisciplinary team that includes experimental and
computational expertise, and your work will focus on collecting and
analyzing Dielectric Relaxation Spectroscopy (DRS) data for lithium ion
electrolytes. This data will be crucial for helping our team better
understand the role of ion dynamics play in controlling ion mobility by
identifying characteristic timescales of ion motion and distinguishing
populations of different solvated and aggregated ionic species. Ultimately,
this improved understanding of ion dynamics will be used to design new
electrolyte materials for next generation energy technologies.
Studying the Role of Dynamics in Electrolytes for Lithium Ion Batteries
"Predicting extreme event statistics for ship motions and loads"
Faculty Advisor: T. Sapsis
Mentor(s):
Contact e-mail: sapsis@mit.edu
Research Area(s):
The idea is to apply machine learning algorithms to bridge the gap between i) expensive, high-fidelity numerical tools and ii) low-cost, low-fidelity models, which simulate motions and loads of ships in random seas. Extreme event analysis involve probabilities of events that happen very rarely and therefore require a huge number of simulations. On the other hand, high-fidelity numerical tools are very expensive to run for long periods. This project will aim to machine learn a connection or a map between the results of the two numerical methods. If such map is known then one can utilize very long runs of the low-fidelity model in order to reconstruct simulations as if these were produced by the high-fidelity model. Students are expected to have familiarity with Python and be eager to learn basic machine-learning techniques, as well as how to run high-fidelity codes describing ship motions and loads. Predicting extreme event statistics for ship motions and loads
"UAVs/Drones for Equitable Climate Change Adaptation: Participatory Risk Management through Landslide and Debris Flow Monitoring in Mocoa, Colombia"
Faculty Advisor: Justin Solomon and John Fernandez
Mentor(s): Norhan Bayomi
Contact e-mail: nourhan@mit.edu
Research Area(s):
Colombia has experienced significant landslide disasters, including the 2017 landslide in Mocoa that killed over 300 people and the 1985 Armero disaster which killed more than 20,000 people, being the country’s most deadly landslide disaster to date. Colombia has made progress in recent years in terms of policies and planning instruments for disaster risk management, exemplified by its 2015-2025 National Disaster Risk Management Plan which is directly linked to its national legislation and which aims to reduce disaster mortality by 2025 while also reducing the negative impacts on livelihoods from recurring climate-related hazards such as floods, droughts, landslides and forest fires. However, the implementation of these measures is slow at local level. Surrounded by mountainous rainforest and with six rivers running through and around Mocoa (City in Colombia), the town’s topography and location makes it prone to mudslides and flash floods. This danger is exacerbated as a result of deforestation in the upper mountainous region for cattle ranching and other agriculture, removing protection against flooding and landslides. This project aims to provide an effective and robust landslide monitoring system that combines data collected by Unmanned Aerial Vehicles (UAVs) to develop an innovative algorithm using machine learning and artificial intelligence to model and predict landslide probability that establishes key factors in land displacement and risk mitigation scenarios, while strengthening the capacity of local authorities and communities to participate in risk management. Moreover, a publicly accessible decision support web tool will be developed for interactive visualization of results for scenario planning, risk assessment, and to inform local adaptation strategies.

Required Skills: Computer Vision and Deep Learning using Neural Networks interest with special emphasis on geometric data

Deliverables: An algorithm for land slide probability using machine learning methods to analyze weekly flights data and identify changes in land elevation profile.
UAVs/Drones for Equitable Climate Change Adaptation: Participatory Risk Management through Landslide and Debris Flow Monitoring in Mocoa, Colombia
"Flame synthesis of catalysts for pollutant reduction"
Faculty Advisor: Sili Deng
Mentor(s):
Contact e-mail: silideng@mit.edu
Research Area(s):
Flame not only generates unhealthy smoke or soot but can also generate useful catalysts for water treatment, energy conversion, and pollutant degradation. In this project, we are going to build a burner and design the flame conditions to synthesize metal oxide catalysts for hazardous gas reduction. Your job: You will work closely with the professor and a graduate student to (1) design and build a burner; (2) optimize the flame conditions for catalysts production; (3) characterize the morphology and performance of the catalysts. This project is interdisciplinary, combining thermosciences, materials sciences, and electrochemistry. You will have a lot of fun playing with fire and making good use of it. Flame synthesis of catalysts for pollutant reduction
"Ellipsoid Packing and Applications to Chromosome Organization"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): BioEECS, Computational Biology, Control and Decision Systems, Inference, Machine Learning, Numerical Methods
The spatial organization of the genetic material in the cell nucleus is known to be important for gene regulation. During most of the cell cycle each chromosome occupies a roughly ellipsoidal domain in the cell nucleus. Hence, the spatial organization of chromosomes can be modeled as an ellipsoid packing problem: 46 ellipsoids of a given size and shape (the chromosomes) should be packed into an ellipsoidal container (the cell nucleus) so as to minimize their overlap under various constraints. The goal of this project is to develop scalable optimization algorithms to determine the packing of the genome inside the cell nucleus and validate the predictions based on imaging data. Ellipsoid Packing and Applications to Chromosome Organization
"Causal Inference and Reinforcement Learning (RL)"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): Computational Biology, Control and Decision Systems, Inference, Machine Learning
In many applications the end goal of causal inference is not necessarily to learn the underlying causal system but to infer the best interventions in order to push the underlying system towards a particular desired state. This is the case for example when studying reprogramming, where the goal is to determine the best interventions (e.g. over-expression of particular transcription factors) to push a differentiated cell towards the stem cell state. In this project, the goal is to build on methods in RL and in causal inference to obtain methods for selecting the best interventions in order to push the system towards a desired state. Causal Inference and Reinforcement Learning (RL)
"Learning Causal Graphs and Applications to Gene Regulation"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): Computational Biology, Control and Decision Systems, Inference, Machine Learning
Causal inference is a cornerstone of scientific discovery because it asks `why?'. Most methods for learning causal directed graphs assume that the underlying graph is a DAG, i.e., that it does not contain any directed cycles. However, feedback loops in biological networks are not only common but also crucial features. In addition, many causal inference algorithms do not allow for imposing prior knowledge on the directed graph or cannot be applied to large networks. In this project, the goal is to develop causal inference algorithms that can overcome these limitations and can be applied to infer gene regulatory networks. These networks have about 20'000 nodes, but there is a lot of prior information on the network coming from knock-out experiments. In order to increase the power of the methodology, it is important to be able to use this prior information. What statistical guarantees can be obtained and what is the computational trade-off? How well does the algorithm perform on simulations? Does it provide meaningful gene regulatory networks when applied to real biological data? Learning Causal Graphs and Applications to Gene Regulation
"Biomimetic MEMS Sensors"
Faculty Advisor: Michael Triantafyllou
Mentor(s): Tom Consi
Contact e-mail: mistetri@mit.edu
Research Area(s):
We are developing sensors inspired by the biological sensors that fish and other marine animals use as feedback for the control swimming and maneuvering. Such sensors will be used in the control of biomimetic marine robots and other high-performance marine vehicles. They also have potential for use in new and novel hand-held environmental sensing devices. We have recently developed three of biomimetic sensors: a flow sensor analogous to the lateral line of fish, a flow sensor modeled on seal whiskers, and a chemical sensor based on the olfactory system of sharks. The next phase of the research is to develop miniature, stable, and low-power circuits to amplify and filter the signals from the sensors. We must also develop a communication and control network that will read an array of sensors and transfer the information to the robot’s control computer. These are the goals of the super-UROP – to develop miniature, stable and robust electronics that will permit the reliable operation of our biomimetic sensor arrays in a marine robot.
We are looking for a student who has excellent theoretical knowledge and hands-on experience in analog circuit design and signal processing. Knowledge of low-power design and embedded microcontrollers is also desirable.
"Mechano-responsive, color-changing materials: design, scalable manufacture, and applications"
Faculty Advisor: Mathias Kolle
Mentor(s): Ben Miller
Contact e-mail: bmill@mit.edu, mkolle@mit.edu
Research Area(s):
This SuperUROP project is focused on exploring the utility of color-tunable photonic sheets in multiple application areas. We have recently developed a new technique for creating elastic materials that change color when deformed. This technique is versatile, scalable, and low-cost, enabling a variety of potential applications and devices. One particular area of interest is combining these materials with simple imaging systems, to create pressure and strain sensors that can be used for human-computer interaction and robotic sensing. We're proposing a project within this space which will involve prototyping physical devices, image processing/computer vision, mechanical modeling, and demonstrating example applications. The project will have a good balance of hands-on making, programming, and theory, and you’ll have an opportunity to explore other application areas depending on your scientific interest. Mechano-responsive, color-changing materials: design, scalable manufacture, and applications
"A Novel Applicator to Increase Condom Utilization"
Faculty Advisor: Ellen Roche
Mentor(s): Ravi Rasinglam
Contact e-mail: etr@mit.edu
Research Area(s):
Rates of unintended pregnancies and sexually transmitted infections (STIs) are increasing and especially affect vulnerable segments of society. Condoms are the only reversible contraceptive option for men, and the only method for STI prevention in sexually active individuals. When used appropriately, condoms are effective in preventing STIs and pregnancy, yet condom use represents only 7.7% of contraception methods globally. This is in part due to negative perceptions associated with condoms such as loss of intimacy, trust and promiscuity. These perceptions are intertwined with complaints regarding how condoms feel for both partners and, for some men, complaints about maintaining an erection. All of these factors contribute to condom use errors, including not placing the condom before intercourse, not removing after ejaculation, and damage to the integrity of the barrier during placement. While many developers have focused on material science to alter the fit and feel of condoms, we believe that to address the major drivers of insufficient use requires a design that focuses on behavioral considerations and the user experience. Our hypothesis is that a novel applicator method that facilitates efficient and effective application and penile stimulation during condom placement is a viable approach to increase condom use. A Novel Applicator to Increase Condom Utilization
"A Novel Treatment for Obstructive Sleep Apnea"
Faculty Advisor: Ellen Roche
Mentor(s): Tarsha Ward
Contact e-mail: etr@mit.edu
Research Area(s):
Obstructive sleep apnea (OSA) is a common, chronic condition characterized by recurrent collapse of the upper airway during sleep (apnea). The reoccurring events lead to frequent arousals that cause daytime fatigue and drowsiness. OSA has been associated with increased heart attack, stroke and premature death. The current ‘gold-standard’ treatment is continuous positive airway pressure (CPAP, pictured) that requires a tight-fitting mask and uses air forced under high pressure to ‘stent’ the throat and airway open. While able to reduce apnea, most patients do not tolerate it and remain untreated. Oral prostheses are an alternative treatment but also are uncomfortable to wear. Our approach leverages advances in oral prostheses and a unique system to stabilize the tissues of the throat to prevent apnea yet remain comfortable and effective. Within this project the student will use innovative methods of 3D design and printing to facilitate manufacturing of a small medical device. The project will include material science investigation, and collection and analysis of sleep data. We have a multi-disciplinary team of scientists and clinicians with a range of expertise that will support the student in the project. Students will also interact with patients and participate in customer discovery for the project. A Novel Treatment for Obstructive Sleep Apnea
"Deep Sea Mining"
Faculty Advisor: Thomas Peacock
Mentor(s):
Contact e-mail: tomp@mit.edu
Research Area(s):
A major consideration for the sustainability of the ocean, and indeed for the sustainability of the planet, is that in the next decade deep-sea polymetallic nodule mining operations are expected to commence. These nodule resources, at depths of 4,500 m in the ocean abyssal plains, contain vast reserves of Nickel, Copper, and Cobalt; in the eastern Pacific Ocean alone there is over six times more Cobalt and three times more Nickel than current global land-based reserves. The driver is the growing demand for these metals to support the development of a low-carbon, circular, global economy, which demands power storage, electric vehicles and clean energy. There are, however, significant concerns about the impact that deep-sea nodule mining operations will have on the abyssal environment given the current lack of data. This SuperUROP research project will involve working with postdocs and graduate students in the Environmental Dynamics Lab (ENDLab) to study the nature of deep-sea mining sediment plumes. Depending on the interests and skills of the student, it will involve some combination of lab experiments, modeling, instrumentation development and processing of data from a major field program scheduled for Fall 2020.
"Computational Linguistics and Language Education"
Faculty Advisor: Michael Cuthbert
Mentor(s):
Contact e-mail: cuthbert@mit.edu, digitalhumanities@mit.edu
Research Area(s):
A SuperUROP in the Digital Humanities Lab during AY20-21 will work on using machine learning and methods from computational linguistics and from frontend web development to create an application to assist language teachers in their classroom teaching. They will help computers understand and implement effective techniques in language acquisition that can transform language teaching at MIT and beyond.

SuperUROPs in the Digital Humanities Lab will work closely with the DH Faculty Director and our Digital Humanities Faculty Fellows to engage with the subject material from the humanities, and learn essential skills of communicating along interdisciplinary boundaries. They will also work with the Digital Humanities Technical Director to create novel technical infrastructure for the project, the creation of which will itself be publishable research.
Computational Linguistics and Language Education
"Photographic Analysis of Political Leanings"
Faculty Advisor: Michael Cuthbert
Mentor(s):
Contact e-mail: cuthbert@mit.edu, digitalhumanities@mit.edu
Research Area(s):
A SuperUROP in the Digital Humanities Lab during AY20-21 will work on using machine learning and computer vision to analyze and categorize photographs by French photographers working in Maoist China. We will explore whether machine learning techniques can help us discover traces of political affiliation in these photos, where the sheer number of photographs make it difficult or impossible to analyze by hand.

SuperUROPs in the Digital Humanities Lab will work closely with our DH Faculty Director and Digital Humanities Faculty Fellows to engage with the subject material from the humanities, and learn essential skills of communicating along interdisciplinary boundaries. They will also work with the Digital Humanities Technical Director to create novel technical infrastructure for the project, the creation of which will itself be publishable research
Photographic Analysis of Political Leanings
"Zorro: A System to Fairly Collect and Price Consumer Data"
Faculty Advisor: Devavrat Shah
Mentor(s): Anish Agarwal
Contact e-mail: anish90@mit.edu
Research Area(s): Computer Systems, Inference, Machine Learning, Signals and Systems
Personal data is a key ingredient in showing web users targeted ads - the economic backbone of the internet (a $100 Billion dollar a year industry). Still, there are two major inefficiencies in how such data is bought and sold online: Consumers do not get to: (a) decide what types of ads they get to see; (b) decide what personal information is released; (c) get paid for this privacy loss. Advertisers purchase user data without knowing how much value it provides, leading to somewhat arbitrary data contracts. Zorro, is a web-based software platform that aims to rectify the aforementioned two problems. Zorro: A System to Fairly Collect and Price Consumer Data
"Transformation tolerance of machine-based face recognition systems"
Faculty Advisor: Pawan Sinha
Mentor(s): Kyle Keane
Contact e-mail: psinha@mit.edu, kkeane@mit.edu
Research Area(s):
Several companies with substantial investment in AI technologies, such as Google, Facebook and Baidu, have declared success on the profoundly important task of face recognition. (“Facebook’s facial recognition software is now as accurate as the human brain, but what now?” asks one 2014 headline from ExtremeTech.) If true, this would be a remarkable accomplishment. Face recognition is widely acknowledged to be a very challenging vision problem, given the infinitely many variations possible in face images. Remarkably, humans excel at this task, robustly recognizing faces even under dramatic degradations of images. Machine vision systems have typically been tested under very benign conditions of high-quality images and restricted pose and lighting parameters. It is unclear how well they perform in more challenging settings.
Over a decade ago, researchers in the Sinha lab at MIT wrote a paper titled “Face recognition by humans: 19 results all computer vision researchers should know about.” Published in Proceedings of the IEEE, this paper had significant impact on the field, and has been cited over 800 times. In characterizing human performance along multiple transformational dimensions, it laid out a set of benchmarks against which to evaluate other face recognition systems. With this foundation in place, our goal in this project is to examine how well state of the art deep network-based face recognition systems perform on these yardsticks, relative to humans.
The results will be very instructive and will reveal the vulnerabilities of state of the art systems. This knowledge will, on the one hand, guide a more careful deployment of these systems to ensure that they are not forced to confront the kinds of settings that prove difficult for them. On the other hand, our results will point the way towards avenues for improvement of the current systems, by defining the frontiers of their performance.
In concrete terms, the project will involve extensive computational experimentation with several deep networks on large databases of face images transformed along a variety of dimensions, as defined in the IEEE paper. The networks will be trained on augmented and unaugmented face databases, and one of our goals will be to determine the kind of training set augmentation that yields the best test performance. We may also undertake studies of human performance if existing data are judged to be inadequate for a fair comparison between machine and human observers. The project will, therefore, span the domains of human and machine perception and yield results that enhance our understanding of both.
Transformation tolerance of machine-based face recognition systems
"Identifying viable livers for transplant"
Faculty Advisor: Nick Roy
Mentor(s): Katherine Gallagher
Contact e-mail: nickroy@mit.edu, kvg0@mit.edu
Research Area(s):
The Quest for Intelligence, working with Massachusetts General Hospital, is interested in leveraging computer vision techniques to identify viable liver transplants from biopsy images. Currently, there are not enough livers available to meet demand: approximately 14,000 transplant candidates are waitlisted annually versus 8,000 livers transplanted. A number of approaches have been suggested to address this deficit, including 1) using livers with acceptably high fat content and 2) extending the transplant time window by freezing, then rehabilitating livers.

Using a dataset of liver biopsy images and accompanying pathologist evaluations, the SuperUROP will research appropriate techniques/architectures for a model (or models) that can accurately, reliably, and transparently 1) identify which livers have an acceptable fat content and/or 2) are otherwise viable after undergoing a freezing/rehabilitation process. The SuperUROP will also develop the appropriate codebase to train, test, and deploy the model(s).
Identifying viable livers for transplant
"Automated Gaze Tracking for Early Childhood Cognition Research"
Faculty Advisor: Nick Roy
Mentor(s): Katherine Gallagher
Contact e-mail: nickroy@mit.edu, kvg0@mit.edu
Research Area(s):
The Quest for Intelligence, working with the Early Childhood Cognition Lab, is interested in facilitating human cognitive development research via an automated gaze tracking system for infants. Many cognition studies collect behavioral data by recording a subject watching a left/right alternating stimulus, then interpreting the subject’s gaze target as a proxy for attention. These recorded videos of subjects watching stimuli are currently hand-coded at each timestamp, a laborious process that we are looking to automate with a computer vision model.

Using an existing dataset of hand-coded videos, the SuperUROP will research appropriate techniques/architectures for this model and develop the appropriate codebase to train, test, and deploy the model.
Automated Gaze Tracking for Early Childhood Cognition Research
"Physiological Time Series Dynamics and Treatment Response Modeling Using Machine Learning"
Faculty Advisor: Roger Mark
Mentor(s): Li-wei Lehman
Contact e-mail: lilehman@mit.edu
Research Area(s): BioEECS, Machine Learning, Signals and Systems
Project Description: Cardiovascular variables such as heart rate and blood pressure are regulated by an underlying control system, and therefore, the time series of these vital signs exhibit rich dynamical patterns of interaction in response to external perturbations (e.g., drug administration), as well as pathological states (e.g., onset of sepsis). A question of interest is whether complex dynamical patterns exhibited across a heterogeneous patient cohort can be used for prognosis of patients’ health and progress. This project aims to apply machine learning techniques to model time-varying changes in multivariate physiological/clinical time series for informed clinical treatment decision making. Example machine learning techniques we will consider include switching state space models, Gaussian processes, and recurrent neural networks. This project will involve working with real-world physiological waveforms/signals collected from patient monitors as well as longitudinal clinical time series data from electronic health records (EHRs). Relevant URL: Link

Prerequisite: The candidate should have experience in machine learning. Knowledge and experience in one or more of the following areas would be desirable: signal processing, probabilistic graphical models, state space models, Gaussian processes, dynamical systems, and deep learning.

Applicants should submit their resume and research interests via an email directly to Dr. Li-wei Lehman, lilehman@mit.edu
Physiological Time Series Dynamics and Treatment Response Modeling Using Machine Learning
"Using Machine Learning to Reduce False Arrhythmia Alarms in the Intensive Care Units"
Faculty Advisor: Roger Mark
Mentor(s): Li-wei Lehman
Contact e-mail: lilehman@mit.edu
Research Area(s): BioEECS, Machine Learning
False arrhythmia alarm rates in Intensive Care Units (ICUs) may be as high as 89%, leading to disruption of care and “alarm fatigue.” The aim of this study is to explore the use of machine learning techniques to address the problems of reducing false arrhythmia alarms (e.g., asystole, extreme bradycardia, extreme tachycardia, ventricular tachycardia, and ventricular flutter/fibrillation) in the intensive care units (ICUs) and distinguishing clinically important events from noise and artifact. We aim to develop techniques based on multiple physiological waveforms measured in standard clinical monitoring (ECG, arterial blood pressure, and photoplethysmogram). Relevant URL: Link

Prerequisite: The candidate should have experience in machine learning. Knowledge and experience in deep learning or other advanced machine learning techniques for waveform/signal analysis would be helpful.
Applicants should submit their resume and research interests via an email to Dr. Li-wei Lehman, lilehman@mit.edu.
Using Machine Learning to Reduce False Arrhythmia Alarms in the Intensive Care Units
"Control and computing platforms  for complex dynamical systems: The case of future  aircrafts and space vehicles"
Faculty Advisor: Marija Ilic
Mentor(s):
Contact e-mail: ilic@mit.edu
Research Area(s): Applied Physics, Control and Decision Systems, Energy, Power, Electromagnetics, Numerical Methods
Unconventional designs and challenging missions require modeling and control of dynamics for ensuring feasibility, stability and desired performance metrics
There exist many non-unique control designs; closed-loop dynamics result in qualitatively different performance
Aircraft electrification offers potential cooperative control within turbo-electric distributed propulsion
Control/storage potentially effective for stable engine performance—no engine stall/surge; diverse impact on aircraft design
To explore these opportunities—view aircraft as a system-of-systems (SoS) comprising dynamic systems/subsystems


Build n the work started in

Ilic, M., & Jaddivada, R. (2019). Exergy/energy dynamics-based integrative modeling
and control for difficult hybrid aircraft missions. In AIAA Propulsion and Energy 2019 Forum(p. 4501).

Opportunities to collaborate with NASA.
Control and computing platforms  for complex dynamical systems: The case of future  aircrafts and space vehicles
"Control co-design for clean and resilient electric energy systems"
Faculty Advisor: Marija Ilic
Mentor(s):
Contact e-mail: ilic@mit.edu
Research Area(s): Applied Physics, Computer Architecture, Control and Decision Systems, Energy, Power, Electromagnetics, Numerical Methods
In this project we utilize the concept of control co-design as a new method for integrating new technologies (storage, clean renewable resources, adaptive loads) and designing their embedded control, sensors, and computing to make the most out of available energy while meeting diverse needs.

For general reading on control co-design, see
Garcia-Sanz M. (2019). Control Co-Design: an engineering game changer.
Advanced Control for Applications, Vol. 1, Num. 1, Wiley

The question in this project is how to utilize this concept to enable clean and high quality electric energy services.

Two specific examples:

1) Can we make MIT power consumer behave adaptively and help the other electric energy users in Cambirdge?

2) Can we use power electronics switching to support stable services without deploying excessive amounts of expensive storage?

3) What does control co-design mean for a stand-alone island, see
Ilic, Marija, Le Xie, and Qixing Liu, eds. Engineering IT-enabled sustainable electricity services: the tale of two low-cost green Azores Islands. Vol. 30. Springer Science & Business Media, 2013.
Control co-design for clean and resilient electric energy systems
"Battery-Free Subsea IoT: Oceans & Climate"
Faculty Advisor: Fadel Adib
Mentor(s):
Contact e-mail: fadel@mit.edu
Research Area(s): Communications, Computer Networks, Signals and Systems
The project focuses on developing battery-free sensor for the subsea IoT. It is based on award-winning technology from our lab, which has demonstrated the ability to communicate with sensors without requiring them to have any batteries.

Here is a link to a video describing the technology: Link
Battery-Free Subsea IoT: Oceans & Climate
"Advanced Heat Transfer Microfluidics via Generative Design-Enabled Additive Manufacturing"
Faculty Advisor: Luis Fernando Velasquez-Garcia
Mentor(s): Javier Izquierdo
Contact e-mail: lfvelasq@mit.edu, javieriz@mit.edu
Research Area(s): Applied Physics, Energy, Power, Electromagnetics, Machine Learning, Materials, Devices and Photonics, Nanotechnology, Numerical Methods
The aim of this project is to explore new design methodologies based on Artificial Intelligence using the characteristics and advantages of additive manufacturing for the generation of microsystems that are applicable to the new electro-mobility industry such as heat sinks for GPUs and CPUs, sensors and electrical devices such power electronics or motors. The work can be theoretical (implementing AI approach), experimental (3d printing devices, testing them and gathering data), or a combination of both. We look for very motivated people, passionate about high-tech with previous exposure to one or more fields: generative design, 3d printing, microfluidics, experimental work, heat transfer (exposure to only one field is fine). Advanced Heat Transfer Microfluidics via Generative Design-Enabled Additive Manufacturing
"Differentiable 3D rendering for deep learning and realistic graphics."
Faculty Advisor: Fredo Durand
Mentor(s):
Contact e-mail: fredo@mit.edu
Research Area(s): Computer Graphics and Vision
Our group has developed algorithms to compute the derivatives of a 2D image with respects to parameters that describe the 3D scene (geometry, camera location, lighting, material, texture). This super urop explores applications to deep learning and inverse 3D problems. Differentiable 3D rendering for deep learning and realistic graphics.
"Computational Photography"
Faculty Advisor: Fredo Durand
Mentor(s):
Contact e-mail: fredo@mit.edu
Research Area(s): Computer Graphics and Vision
This project deals with algorithms and deep learning methods to enhance the quality of photos, in particular for low-quality cell phone devices. Computational Photography
"Joint design of optics and computation for cell phone cameras"
Faculty Advisor: Fredo Durand
Mentor(s):
Contact e-mail: fredo@mit.edu
Research Area(s): Computer Graphics and Vision
We seek to jointly design optics and computation. For this, we implement an optics simulator together with a software image reconstruction pipeline and apply backpropagation (similar to deep learning) and numerical optimization jointly to the hardware and software. This will be applied to regular camera phones, but also to light field (plenoptic) design to improve cell phone imaging, and also to 3D acquisition systems. Joint design of optics and computation for cell phone cameras
"Visualizing How Readers Read and Interpret Literature"
Faculty Advisor: Sandy Alexandre
Mentor(s):
Contact e-mail: alexandy@mit.edu
Research Area(s):
Could seeing a one-page visual analysis of how you have read a piece of writing help you not only better understand how you read and find meaning, in general, but also help you write about that piece of writing more clearly, more easily, and more enjoyably than you otherwise would have without such a visual aid? How can we visually trace or track the process of a person's digital reading experience in such a way that benefits that person when they eventually have to write about what they have read?

This research project aims to answer these questions. The ideal research assistant would need to be skilled in visualizing data and also have some familiarity with NLTK. Some of the legwork has already been done for this project (the codebase already exists, for example), but I am also looking for research assistants who have the capacity to polish the user interface of the current version of the website I have for the project. Some experience designing visually appealing and user-friendly websites is a huge plus!
Visualizing How Readers Read and Interpret Literature
"Model-Agnostic Meta-Learning: From Supervised to Reinforcement Learning"
Faculty Advisor: Asuman Ozdaglar
Mentor(s): Alireza Fallah
Contact e-mail: asuman@mit.edu, afallah@mit.edu
Research Area(s): Machine Learning
In a large fraction of artificial intelligence problems, ranging from robotics to image classification and pattern recognition, the goal is to design systems that use prior experience and knowledge to learn new skills more efficiently. Meta-learning formalizes this goal by using data from previous tasks to learn update rules or model parameters that can be fine-tuned to perform well on new tasks with a small amount of data.

The goal of this project is to develop meta-learning algorithms for both supervised and reinforcement learning applications and measure their performance empirically and in comparison with other existing methods in the literature. In particular, over the past few months, we have been working on designing novel and efficient meta-learning algorithms [1,2] and characterizing their convergence properties, and we are looking for SuperUROP students who could implement these algorithms for several datas sets and tasks. The resulting experiments would guide developing new methods with better performance. The project could also be extended to understating further theoretical aspects of meta-learning algorithms.

Experience with Python and Pytorch/Tensorflow is highly desired. Familiarity with basic optimization algorithms such as garden descent is required.

[1] A. Fallah, A. Mokhtari, A. Ozdaglar, On the convergence Theory of Gradient-Based Model-Agnostic Meta-Learning Algorithms, To appear in AISTATS 2020,
arXiv preprint arXiv: 1908.10400, 2019.

[2] A. Fallah, A. Mokhtari, A. Ozdaglar, Provably Convergent Policy Gradient Methods for Mode-Agnostic Meta-Reinforcement Learning, arXiv preprint arXiv:
2002.05135, 2020.
Model-Agnostic Meta-Learning: From Supervised to Reinforcement  Learning
"Social Media and the Syrian conflict"
Faculty Advisor: Fotini Christia
Mentor(s): Kiran Garimella IDSS
Contact e-mail: cfotini@mit.edu
Research Area(s):
In many areas of the world, messaging apps such as What's App and Telegram are responsible for the largest percentage of Internet traffic. Aggregating these messages and looking at trends in rhetoric and image sharing between groups and across time can help us understand political and social trends. We are using these techniques to analyze the political and refugee situation in the northern part of Syria. This region on the border with Turkey is particularly volatile, as the Assad regime has less control and Turkey has been using military force to assert influence.

Possible project areas to contribute to include text analysis of Telegram messages, image analysis, web scraping of Twitter and text analysis of Twitter messages. Arabic language skill is a plus for text analysis if interested (but definitely not required).
Social Media and the Syrian conflict
"A New Web Architecture"
Faculty Advisor: Daniel Jackson
Mentor(s):
Contact e-mail: dnj@mit.edu
Research Area(s): Human Computer Interaction, Programming Languages (incl software eng)
We are working on a new web ecosystem where website developers expose more structured data in web clients, to support easier modification by end users, and peer to peer sharing. But Wildcard is also pragmatically designed to work with the existing websites of today, using adapters that map between the website and a spreadsheet like view.

We have implemented this technique in a prototype browser extension called Wildcard. Through concrete examples, we demonstrate that Wildcard can support useful customizations—ranging from sorting lists of search results to showing related data from web APIs—on top of existing websites.

For more info see
Link
Link
A New Web Architecture
"Certified Control for Autonomous Cars"
Faculty Advisor: Daniel Jackson
Mentor(s):
Contact e-mail: dnj@mit.edu
Research Area(s): Programming Languages (incl software eng), Robotics, Systems (incl OS, databases, computer security)
Certified control is a new architecture for autonomous cars that offers the possibility of a small, verifiable trusted base without preventing the use of complex machine-learning algorithms for perception and control.

The key idea is to exploit the classic gap between the high cost of finding a solution to a problem and the much lower cost of checking that solution. The main controller plays the role of the solver, analyzing the scene and determining an appropriate next step, and the certifier plays the role of the checker, ensuring that the proposed step is safe.

To make this check possible, the main controller constructs a certificate that captures its analysis of the situation along with the proposed action. The main controller is thus excluded from the trusted base: when it works correctly, the certifier endorses its commands; and when it fails, the certifier will reject the commands and a simpler controller will bring the car to a safe stop. We have designed an architecture that embodies this idea, and demonstrated it in simulation and in a racecar.

We are looking for SuperUROPs who have taken coursework in robotics or self-driving cars, and have strong programming skills in Python or Java. Machine learning knowledge is not required.

For more info see
Link
Certified Control for Autonomous Cars
"2-D Bistable Optical Array"
Faculty Advisor: Cardinal Warde
Mentor(s): none
Contact e-mail: warde@mit.edu
Research Area(s): Circuits, Machine Learning, Materials, Devices and Photonics
The Photonic Systems Group is developing a low-power Compact Opto-electronic Integrated Neural (COIN) network co-processor for applications such as image recognition and classification, and sensor fusion. The COIN, which is inspired by biology, employs arrays of photodetectors and low-power thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. We are exploring the development of such a neural network system at the discrete component level first, before committing to the chip-based integrated version. One of the key hardware subsystems is a dense 2-D array of bistable optical devices.

This project involves: (1) design, modeling and assembly of the low-power components and the associated optical bistable subsystem,and (2) assembly and testing of a hybrid optical bistable array. Follow-on work could focus on contributing to the design and fabrication of the final compact VLSI integrated-circuit version of the COIN.
2-D Bistable Optical Array
"Assembly of Hybrid COIN"
Faculty Advisor: Cardinal Warde
Mentor(s): none
Contact e-mail: warde@mit.edu
Research Area(s): Circuits, Computer Architecture, Machine Learning, Materials, Devices and Photonics, Signals and Systems
The Photonic Systems Group is developing a low-power Compact Opto-electronic Integrated Neural (COIN) network co-processor for applications such as image recognition and classification, and sensor fusion. The COIN, which is inspired by biology, employs arrays of photodetectors and low-power thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. We are exploring the development of such a neural network system at the discrete component level first, before committing to the chip-based integrated version.

This project involves: (1) design and modeling of the overall neural network system, and (2) assembly and testing of the system with real-world input after weight training. Follow-on work could focus on contributing to the design and fabrication of the final compact VLSI integrated-circuit version of the COIN.
"Training Algorithms for the COIN"
Faculty Advisor: Cardinal Warde
Mentor(s): none
Contact e-mail: warde@mit.edu
Research Area(s): Computer Systems, Machine Learning, Signals and Systems, Theoretical Computer Science
The Photonic Systems Group is developing a low-power Compact Opto-electronic Integrated Neural (COIN) network co-processor for applications such as image recognition and classification, and sensor fusion. The COIN, which is inspired by biology, employs arrays of photodetectors and low-power thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. We are exploring the development of such a neural network system at the discrete component level first, before committing to the chip-based integrated version.

This project involves the development and implementation of neural network algorithms to first train a software model of the COIN, prior to training the actual COIN hardware.
Training Algorithms for the COIN
"Making Governments More Transparent: Using Big Data and Experiments to Change Government Behavior"
Faculty Advisor: Daniel Hidalgo
Mentor(s):
Contact e-mail: dhidalgo@mit.edu
Research Area(s):
How transparent are local governments in the US? WHat kinds of pressures are most likely to induce local governments to improve their level of transparency? To answer these questions, I have developed a methodology using machine learning to measure the degree of government transparency in local government websites in the US. This data has revealed wide variation in the degree to which local governments make it easy to observe their internal operations through budgets, meeting minutes, etc. To understand the best way to reduce these disparities in transparency, I am engaging in a set of randomized trials where this information is given to either the governments themselves or to local media organizations.

Students in this UROP will help me improve the machine learning methodology and conduct the experiments. This will involve collecting and checking data, contacting governments and local media sites, and conducting background research.
Making Governments More Transparent: Using Big Data and Experiments to Change Government Behavior
"Spatial Sound Lab for GIRs, Art, Music, and Ethnography"
Faculty Advisor: Ian Condry
Mentor(s):
Contact e-mail: condry@mit.edu
Research Area(s):
The MIT Spatial Sound Lab is a new community production studio for immersive sound projects using a 16.4 speaker system and new computing technology for object-oriented, spatial audio mixing. We are eager to work with students interested in programming new tools and plug-ins for creating live and recorded spatial mixes of music, sound art, sensory ethnography, and teaching modules for the GIRs.
"Privatizing split learning with high communication efficiency for distributed ML"
Faculty Advisor: Ramesh Raskar
Mentor(s): Praneeth Vepakomma
Contact e-mail: vepakom@mit.edu, raskar@media.mit.edu
Research Area(s):
Collaboration and data sharing among individuals, institutions, regional, national and global entities holds key to increasing the amount of available data to be fed as input to data-hungry machine learning models. But centralization of data is hindered through regulatory, ethical, trust, competition and logistical constraints. The SuperUROP will research on pushing the resource and communication efficiency limits of distributed machine learning while catering to challenges of statistical worst-case guarantees on privacy and non-homogeneity of data sets across clients. They are expected to develop methods to learn non-parametric representations that are one-way from an information theoretic point of view. The ultimate goal is to get the most out of the privacy-utility-efficiency-robustness trade-offs through novel algorithms that are theoretically justified. How can all this be done with as few rounds of communication as possible? this makes it suitable to the edge-device ecosystem. The work would involve one or more of distributed machine learning, statistics, deep learning, optimization, split learning and some experimentation with either of PyTorch/TensorFlow/Keras/Matlab/Python/R. Ideal candidate should be able to proactively contribute with periodic updates. Work around adjacently connected problem areas is also encouraged. Both, hands-on experimental projects and/or theoretical opportunity exists. Desired background: Coding fluency/ research mindset. Please contact vepakom@mit.edu and raskar@media.mit.edu to indicate your interest. Please also include your resume.

Pre-requisites:
coding fluency, algorithmic or mathematical mindset encouraged
Privatizing split learning with high communication efficiency for distributed ML
"COVID 19: Procuring Healthcare Interoperability: Achieving High-Quality, Connected, and Person-Centered Care"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Networks, Computer Systems, Inference, Machine Learning, Programming Languages (incl software eng), Systems (incl OS, databases, computer security)
Link

This research project is based on the above report and the critical need for healthcare interoperability based on growing needs and the COVID-19 pandemic.

Despite efforts in the US and in other countries, the problem of transmitting and use of healthcare data across clinics, hospitals, and organizations continues to escalate partly because of new developments in Digital Health. The growing number of medical devices, healthcare algorithms, and the increasing use of telemedicine concepts present a new urgency to the need for surmounting organizational and political boundaries that impede the flow of time-sensitive healthcare data and information. A 57 page paper has been written summarizing major efforts around the world. The research will involve addressing of semantic and other aspects of interoperability across different medical information ecosystems of hospitals, drug manufacturing companies, pharmacies, medical IoT devices, and others in an environment where organizations take deliberate steps to hinder patients' access to their respective medical records. Aspects of this growing problem are highlighted at:
Link
Link
Link

Opportunity to be co-author of a paper in ACM Computing Surveys or another leading journal.
COVID 19: Procuring Healthcare Interoperability: Achieving High-Quality, Connected, and Person-Centered Care
"Addressing Conflicting and Missing Data in Telemedicine and other Electronic Patient Monitoring Environments"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Architecture, Computer Networks, Computer Systems, Control and Decision Systems, Human Computer Interaction, Inference, Machine Learning, Materials, Devices and Photonics, Numerical Methods, Systems (incl OS, databases, computer security)
As envisioned more than 15 years ago by this supervisor and three renowned co-authors who were medical doctors and professors of medicine, healthcare is increasingly using a three-pronged approach involving individuals in proximity of the patient, medical specialists at a distance from the patient, and the use of computer-based analytics. Differences occur frequently when heart rate, blood pressure, and other patient data are collected by both medical personnel and machines, and usually the data collected by medical personnel taken to be the correct one. One study showed that the differences between the two sets of data are highest in the cases of patients who die. The aspect of conflicting data and the aspect of missing data require much deeper analysis as such data drives the treatment of the patient and the likelihood of success of the treatment. This project involves studying of underlying data and creation of computational/statistical algorithms that will allow the medical personnel to efficiently understand the presence and impact of divergent or missing vital sign data for patients. The project is built on the foundations of ideas and activities described at:
Link
and
Link
Addressing Conflicting and Missing Data in Telemedicine and other Electronic Patient Monitoring Environments
"Document Processing with AI & OCR and Neural Network Application in Financial and Medical Realms"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Computer Graphics and Vision, Computer Systems, Human Computer Interaction, Inference, Machine Learning, Natural Language and Speech Processing, Systems (incl OS, databases, computer security)
Previous research by MIT EECS students led to a broad patent on automated reading and processing of bank checks, a concept now used in many countries. This project aims to develop techniques to reduce the human effort involved in transferring information from diverse media to computers, including ID cards (see figure), 1099 and W2 forms, and diverse financial and medical documents. This involves: the detection of the boxes containing the desired information and fields within the box; the separation of material that is in printed and typewritten format versus material in handwritten format so that they can be processed independently with greater automation employed for processing of printed and typewritten material; the classification of the document into one of several categories, each with its own recipe in terms of further processing; and the examination of the document with neural network with respect to potential fraud, security, or other concerns. Artificial Intelligence and Contextual Knowledge will be used to find relevant boxes of information, make preliminary assessment of what the contents of the boxes are, integrate and reconcile the new information with legacy information on computers, and then provide the integrated information for use by financial and medical professionals. Document Processing with AI & OCR and Neural Network Application in Financial and Medical Realms
"Dojo tutor for "second-programming-language" learning"
Faculty Advisor: Rob Miller
Mentor(s):
Contact e-mail: rcm@mit.edu
Research Area(s): Human Computer Interaction, Programming Languages (incl software eng)
Many approaches exist for teaching a first programming language to novice programmers.
The Dojo project aims instead at second-language learning -- how can we add a new tool to a learner's toolbox, whether it's a new programming language, API, programming paradigm, or software engineering technique? And how can we teach it better by building on top of what the learner already knows? SuperUROP projects in this area include: (1) generalizing Java Tutor (currently a Python->Java tutor embedded in Eclipse) to handle other languages and other programming editors; (2) a tutor for higher-level software engineering skills like testing, documentation, refactoring, and debugging; (3) a tutor for data-query languages like SQL, MongoDB, and pandas; (4) phrase-level translation (think Google Translate but for programming) combined with conceptual, pragmatic, and performance recommendations. Good performance in 6.031 is a must; good performance in 6.170 or experience with web development (especially in NodeJS) is a plus.
Dojo tutor for
"Small Ants with Big Wisdom: Consensus in Collective Decision-Making Algorithms in Ant Colonies"
Faculty Advisor: Nancy Lynch
Mentor(s): Jiajia Zhao
Contact e-mail: jiajiaz@mit.edu
Research Area(s): BioEECS, Computational Biology, Theoretical Computer Science
An ant colony is a distributed system with noisy individuals, limited and random connectivity, and no central control. These simple animals with limited intelligence and cognitive capacity can, however, self-organize and present a higher level of intelligence. In this project we focus on decoding their house-hunting process, in which the individuals share a global goal of reaching consensus on a new nest to move to, and moving the entire colony there.

Compared to the distributed algorithms designed by humans, the ants' consensus algorithm achieves a higher level of efficiency and noise tolerance. Therefore, these ant algorithms have drawn a lot of attention from both biologists and engineers. In our work so far, we have developed an initial simulator that allows us to replicate and explain much of the ant colonies' behaviors, but not all. The SuperUROP will explore such extensions as: (1) considering the spatial geometry effect of the candidate nests, (2) considering the effect of changing transition probabilities, (3) using Machine Learning to find optimal but generalizable parameter settings, and (4) drawing a parallel between the collective behavior of ants and other biological systems such as neurons in the brain.

Recommended background: Distributed algorithms, Markov chains and random
processes
Small Ants with Big Wisdom:  Consensus in Collective Decision-Making Algorithms in Ant Colonies
"Low-cost, mass-produced biomedical microfluidics"
Faculty Advisor: Luis Fernando Velasquez-Garcia
Mentor(s):
Contact e-mail: lfvelasq@mit.edu
Research Area(s): Applied Physics, BioEECS, Materials, Devices and Photonics
State-of-the-art microfluidics are artisanally crafted via rubber micromolding —a process that has proven not to be scalable. Via 3D printing, complex (e.g. geometries, actuation) microfluidics can be synthesized; however, current 3D printing technology cannot deliver devices in great quantities and at low cost. In this project, we plan to harness advanced manufacturing to demonstrate the low-cost, scalable fabrication of microfluidics for bio applications. The project has a strong hands-on fabrication component, with likely iteration of the process flows and fundamental characterization of such processes, as well as experimental characterization of simple microfluidic devices, e.g. mixers. Low-cost, mass-produced biomedical microfluidics
"Microplasma sputtering for agile manufacturing of electronics"
Faculty Advisor: Luis Fernando Velasquez-Garcia
Mentor(s): Yosef Kornbluth
Contact e-mail: lfvelasq@mit.edu, ykornblu@mit.edu
Research Area(s): Applied Physics, Materials, Devices and Photonics, Nanotechnology
We are developing a novel atmospheric sputtering system suitable for low-temperature rapid prototyping of electronic components; we have already shown imprints with micron-level minimum feature size and near-bulk electrical conductivity. We are interested in a deeper understanding of the plasma's behavior and its dependence on various process parameters (e.g., gas flow, geometry, power). The challenge lies in the asymmetric geometry of the sputter system, the complexity of plasma physics, and the novelty of atmospheric sputtering approach. The project would focus on simulation of the plasma and the sputtering process, using both COMSOL multiphysics and a more specialized plasma simulator. Experimental verification is a possible addition to this project. Microplasma sputtering for agile manufacturing of electronics
"A Hardware Accelerator for Memory-Efficient Irregular Algorithms"
Faculty Advisor: Daniel Sanchez
Mentor(s): Guowei Zhang, Quan Nguyen
Contact e-mail: zhanggw@mit.edu, qmn@mit.edu, sanchez@csail.mit.edu
Research Area(s): Computer Architecture, Computer Systems
Conventional processors, like CPUs and GPUs, have been designed and optimized for regular applications that operate on structured data, like dense matrices. However, many emerging application domains, such as deep learning, graph analytics, and sparse linear algebra, do not fit this model: these applications perform irregular and unstructured operations on large, sparse data structures, like sparse matrices or tensors. Existing processors cannot execute these applications efficiently, as they incur excessive data movement and cannot use regular processing elements like vector units efficiently.

In this project, you will work with a team of students to design a hardware accelerator for these irregular algorithms. This accelerator uses a new hardware/software interface based on sparse tensors that provides an efficient, unified abstraction for applications in these seemingly disparate domains (e.g., allowing us to build a single accelerator that runs graph and deep learning applications efficiently). Moreover, this accelerator features computational units that make irregular algorithms memory-friendly by restructuring their computation. This approach spends hardware on cheap compute operations to reduce expensive data movement, eliminating the memory traffic bottleneck of conventional architectures.
A Hardware Accelerator for Memory-Efficient Irregular Algorithms
"Unlocking the Potential of Multicore Systems"
Faculty Advisor: Daniel Sanchez
Mentor(s): Victor Ying, Maleen Abeydeera
Contact e-mail: victory@csail.mit.edu, maleen@csail.mit.edu, sanchez@csail.mit.edu
Research Area(s): Computer Architecture, Computer Systems, Programming Languages (incl software eng), Systems (incl OS, databases, computer security)
Current multicores can only exploit a fraction of the parallelism available in applications, and they are very hard to program. We are designing a new type of multicore architecture that tackles both problems. In this architecture, called Swarm ( Link ), programs consist of very short tasks, as small as tens of instructions each. Hardware queues and distributes tasks among cores, reducing the overheads of fine-grain parallelism and allowing many more applications to be parallelized. Moreover, parallelism is implicit: instead of using locks, semaphores, or other error-prone explicit synchronization techniques, programmers simply define an order among tasks. Under the covers, Swarm hardware figures out what order constraints are superfluous and elides them, running most tasks in parallel. As a result, Swarm programs are almost as simple as their sequential counterparts, and at the same time outperform the best parallel programs.

We have already demonstrated Swarm's benefits on challenging applications, like graph analytics and databases, achieving speedups of 50-560x on several parallel algorithms where no prior parallel implementations could beat the sequential ones. But there is much more to do! Given the broad scope of this project, there are many areas where you can contribute, depending on your interests. For example, you can:

- Port challenging parallel applications to Swarm (e.g., circuit and network simulators, SAT solvers, or a program of your choice).

- Improve and generalize the Swarm compiler, T4, which automatically parallelizes sequential C/C++ applications.

- Design a new programming framework to parallelize a broad range of algorithms while retaining the simplicity of sequential programming.

- Help scale Swarm's FPGA implementation, Chronos, which accelerates applications on existing systems.

- Develop new profiling and visualization tools that help programmers understand performance and parallelism bottlenecks.
Unlocking the Potential of Multicore Systems
"Building Trustworthy Agents"
Faculty Advisor: Pulkit Agrawal
Mentor(s):
Contact e-mail: pulkitag@mit.edu
Research Area(s): Machine Learning
Suppose we train artificial agents to perform a particular task. How do we know that agents will solve future tasks in a manner we expect them to? Such questions are particularly important while deploying AI systems such as self-driving cars. In this project, we will study this question. In particular, we will make a game environment in which humans will be tasked to teach agents how to achieve some goals. Because humans will spend time interacting with the agent, they are likely to form models of how the agent is solving a task. The aim of humans is to teach in a manner that agents will generalize in predictable ways. Building Trustworthy Agents
"Investigating Human Preferences via Multi-Sensory Perception"
Faculty Advisor: Pulkit Agrawal
Mentor(s):
Contact e-mail: pulkitag@mit.edu
Research Area(s): Cognitive AI
Information from various sensory modalities is tightly coupled with each other. The sight of strawberries evokes its sweet taste, listening to vroom-vroom sound instantly conjures images of a motorbike or a car! There is plenty of evidence that humans represent the world by integrating information from multiple sensory modalities. This project aims to tease about the exact nature of these representations and explore how predictable are touch, audio, vision and taste signals are from each other. For e.g., is it the case that people who prefer rock music also prefer beer over wine? Investigating Human Preferences via Multi-Sensory Perception
"Ecologically Relevant Self-Supervision"
Faculty Advisor: Pulkit Agrawal
Mentor(s):
Contact e-mail: pulkitag@mit.edu
Research Area(s): Computer Graphics and Vision
Current techniques on self-supervision rely on large internet datasets. However, the data observed by humans is fundamentally different. Instead of observing a a larger variety of objects, we observe a much smaller number of objects, but in much more varied lighting conditions and vantage points. In this project we will explore how different kinds of data influences the generalization and robustness of the learned representations. In particular, does ecologically relevant data (such as one observed by humans) leads to better generalization? Ecologically Relevant Self-Supervision
"Self-Supervised Cloth Manipulation"
Faculty Advisor: Pulkit Agrawal
Mentor(s):
Contact e-mail: pulkitag@mit.edu
Research Area(s): Robotics
Manipulating deformable objects such as clothes is a significant challenge in robotics. This project will investigate how a robot can autonomously collect data to understand various ways in which it can change the configuration of clothes. These models will be then be used to fold clothes and perform other manipulation tasks. We will make use of and advance deep learning, model-based reinforcement learning and imitation learning methods. The most relevant prior work is the lab’s previous project on rope manipulation: Link Self-Supervised Cloth Manipulation
"Encoding point clouds for deep learning"
Faculty Advisor: Justin Solomon
Mentor(s): Dmitriy Smirnov
Contact e-mail: dimkasmir@gmail.com
Research Area(s): Computer Graphics and Vision, Machine Learning, Numerical Methods
As 3D sensors become more and more accurate and commonplace and the amount of 3D data grows, there is an increasing need for methods to automatically analyze this data. While deep learning has proven to be very successful in processing large noisy datasets of images, deep learning on 3D objects such as point clouds is still an emerging field. Recent work has proposed a method to encode point clouds for deep learning pipelines by computing the closest distance from each point cloud to a fixed “base set” of points sampled from the unit sphere. This produces a concise representation, which can better characterize the point cloud than a voxel occupancy grid---albeit at the expense of losing the ability to perform convolutions (which are naturally defined on a voxel grid. This work motivates several exciting research directions and questions: What is special about the unit sphere? Is it possible to design a more effective, perhaps data-specific, algorithm for picking base point sets? Finally, given a base point set, can we use it to define a convolution operator, perhaps borrowing some ideas from the literature of graph neural networks? Encoding point clouds for deep learning
"3D Printing"
Faculty Advisor: Wojciech Matusik
Mentor(s):
Contact e-mail: wojciech@mit.edu
Research Area(s): Computer Graphics and Vision, Human Computer Interaction, Materials, Devices and Photonics, Nanotechnology, Numerical Methods, Robotics
We are looking for an undergraduate student that we can challenge to develop novel applications and processes in multi-material 3D printing. Within the Computational Fabrication Group, we have a number of bespoke 3D printing platforms that can push the boundaries of additive manufacturing. The SuperUROP applicant will be solving high impact problems such as, printing multi-material soft robots that can walk out of the 3D printer after printing, developing AI-driven systems that can design better 3D printing materials than an expert chemist can, or develop new 3D printing technologies that require zero support for any 3D printing. Beside materials and manufacturing processes, we also actively work on applications, such as printed flexible electronics, soft actuators, etc. Applicants should be willing to be hands on and be comfortable in the wet lab. 3D Printing
"Robotics/Learning"
Faculty Advisor: Wojciech Matusik
Mentor(s):
Contact e-mail: wojciech@mit.edu
Research Area(s): Machine Learning, Robotics
We are looking for talented undergraduates to contribute to a few ongoing research projects about robot design/simulation/learning problems: 1) developing computational methods to discover novel robot designs with extremal performances that go beyond what a human expert can achieve. For example, in our previous work, we explored a parametric design space of multicopters to reveal innovative designs with more payload capacity; 2) building differentiable simulations for dynamic, rigid/soft robots (flying vehicles, swimmers, jumpers) and leveraging gradient information to assist planning/control, possibly by combining them with supervised or reinforcement learning; 3) analyzing the geometry or control of robots with optimal performances and developing novel analysis/visualization tools to discover the key component in their design space that makes them optimal. Robotics/Learning
"Intelligent Textiles"
Faculty Advisor: Wojciech Matusik
Mentor(s):
Contact e-mail: wojciech@mit.edu
Research Area(s): Computer Graphics and Vision, Human Computer Interaction, Machine Learning, Materials, Devices and Photonics, Numerical Methods
We are looking for talented undergraduates to contribute to a few ongoing research projects about intelligent wearables/ textile manufacturing/design problems: 1) developing large-scale sensing wearables based on versatile textile manufacturing techniques, such as machine knitting and embroidery; 2) developing a design framework for artists and even novice users to create customized knitted goods with various shapes, styles, patterns, textures, and functionalities; 3) developing a real-to-sim framework to bridge the machine knitted structure and predictive physically-based simulator. Intelligent Textiles
"Cost of Regulating Social Media"
Faculty Advisor: Devavrat Shah
Mentor(s): Sarah Cen
Contact e-mail: shcen@mit.edu
Research Area(s): Cognitive AI, Human Computer Interaction, Inference, Machine Learning
People form judgments and make decisions based on the information they observe. A growing portion of that information is not only provided but specifically curated by social media platforms (SMPs). Content curation is vital to an SMP's financial survival. However, it also has the ability to influence its users' learning and behavior. Indeed, extreme examples, such as the speculated impact of SMPs on elections, have heightened scrutiny on them from regulators and users alike.

In our recent work, we presented a statistical framework for quantifying the effect of SMPs' selected content on user learning.

The goal of this project is two folds: (a) extend the existing framework, and (b) setup experiments to verify the outcomes of our theoretical framework.
Cost of Regulating Social Media
"Machine Learning for Manufacturing"
Faculty Advisor: Duane Boning
Mentor(s): Chris Lang
Contact e-mail: boning@mtl.mit.edu
Research Area(s): Control and Decision Systems, Machine Learning
Manufacturing industries are starting to transform, taking advantage of the exciting promise of machine learning (ML). In this project, we are considering various advanced ML methods, and enabling their application to key challenges in manufacturing. First, modern factories (semiconductor, pharmaceuticals, others) have become highly instrumented, with large numbers of sensors on equipment and throughout the factory. The resulting multivariate time-series data, however, is highly underutilized; we are developing ML time-series approaches to enable manufacturing anomaly detection, predictive maintenance, and monitoring the health of a fleet of tools. Second, models of fabrication processes that utilize both empirical and engineering knowledge and that capture uncertainty are crucial. We are developing Bayesian approaches, including Gaussian Process (GP) ML methods for modeling, and for reinforcement learning and optimization of spatial uniformity and process parameters in manufacturing in the face of this uncertainty. Third, we are also developing ML transfer learning approaches for the visual inspection of product that leverage existing deep neural network (DNN) models and that guide fine-tuning based on limited numbers of defect examples. We seek a student with strong skills in multiple ML areas and in software development, and an interest in applying and translating these methods to real industrial problems. Machine Learning for Manufacturing
"Molecular Devices: Fabrication and Design"
Faculty Advisor: Farnaz Niroui
Mentor(s): none
Contact e-mail: fniroui@mit.edu
Research Area(s): Applied Physics, Materials, Devices and Photonics, Nanotechnology
Molecules with precisely defined intrinsic properties are attractive nanoscale building blocks for the development of novel devices with functionalities that would not be feasible using conventional silicon-based technologies. With dimensions less than 5 nm though, integration of molecules into active devices is a great challenge and has hindered progress in the field. In this project, we will develop techniques that allow us to fabricate arrays of molecular structures with sub-nanometer precision and resolution over large areas. These techniques will provide a platform based on which we can design various molecular devices with emerging applications in electronics, optics and sensing.
"Energy Efficient Security for Biomedical implantable Applications"
Faculty Advisor: Anantha Chandrakasan
Mentor(s): Saurav Maji
Contact e-mail: smaji@mit.edu, anantha@mtl.mit.edu
Research Area(s): BioEECS, Circuits, Energy, Power, Electromagnetics
Implantable medical devices (IMDs) (e.g. cardiac defibrillators, pacemakers, etc) are widely used to improve the quality of lives of patients. However, the wide deployment of these implantable and wearable devices have been limited by their security concerns. Therefore, it is important to ensure the security of these devices, while operating on low-power budgets.
Our group has designed a chip that uses a novel dual-factor authenticated protocol for securing IMD which uses a voluntary human response in addition to cryptographic security. The current system uses touch/tapping based voluntary response. However, the above security protocols could be extended for other biomedical signals as well:
1. Analysis of common biomedical signals appropriate for dual-factor authentication (e.g. EMG/speech signal).
2. Processing of these signals for proper authentication.
3. Energy-efficient implementation of these techniques on embedded platforms.
4. System-level energy modelling of the above protocols.
Prerequisites: Experience with programming in Arduino and Python/C
Energy Efficient Security for Biomedical implantable Applications
"ENGINEERING PROBIOTICS FOR GLUTEN DEGRADATION"
Faculty Advisor: Timothy Lu
Mentor(s): Eugenia Inda
Contact e-mail: inda@mit.edu
Research Area(s): BioEECS, Machine Learning
Celiac disease (CD) is characterized by intestinal inflammation triggered by gliadin, a component of dietary gluten. Oral administration of proteases that can rapidly degrade gliadin in the gut has been proposed as a treatment for CD; however, no protease has been shown to specifically reduce the immunogenic gliadin content, in gastrointestinal conditions, to below the threshold shown to be toxic for celiac patients. Prolyl endopeptidases (PEP) are able to degrade immunotoxic peptides responsible for CD. The goal of this project is to engineer probiotic strains to deliver PEP in the gut as a treatment for CD. From bioinformatics analyses we identified PEP as targets for protein engineering. We are generating an initial library of proteases by introducing mutations around the active site (informed by variants present in other family members) and then we test the performance of all the variants experimentally. Once the initial data is collected, we can use a Machine Learning approach to design subsequent libraries for design iteration. Protein expression will be optimized by testing libraries of promoters and signal peptides for secretion. Several bioengineering and machine learning techniques will be applied in this project to achieve total gluten degradation in the gut. ENGINEERING PROBIOTICS FOR GLUTEN DEGRADATION
"Feedback Control for Robot Manipulation"
Faculty Advisor: Russ Tedrake
Mentor(s): none
Contact e-mail: russt@mit.edu
Research Area(s): Control and Decision Systems
Despite it's prevalence in almost every other field of systems engineering, shockingly almost no state of the art systems for robot manipulation actually use feedback control (because we lack the basic theory). The MIT Robot Locomotion group is pursuing fundamental research in this direction, and applying it to physical robots.

The project will involve (convex/mixed-integer) optimization, some systems theory, and lots of robots -- mostly programmed in python. We don't expect you to know all of that coming in, but hope you're excited about them.

The project may also involve robot perception systems -- developing feedback directly from pixels is a major intellectual theme of our current work.
"Manipulation of Soft Objects"
Faculty Advisor: Russ Tedrake
Mentor(s): none
Contact e-mail: russt@mit.edu
Research Area(s): Control and Decision Systems
The MIT Robot Locomotion Group has been making progress in connecting machine learning perception with robotics planning and control to enable an increasingly diverse set of manipulation capabilities for robotics. (e.g. Link )

We are now working to extend this work to deal with increasingly complex systems -- using machine learning perception, potentially learning intuitive physics, and rigorous dynamics and control + lots of robots.
"Verifying Autonomous Systems with Perception in the Loop"
Faculty Advisor: Russ Tedrake
Mentor(s): none
Contact e-mail: russt@mit.edu
Research Area(s): Control and Decision Systems
State-of-the-art perception systems today are (mostly) based on deep learning. If we aim to make reliable engineering systems using these components (and especially for safety-critical systems), we need to improve our understanding of how these components perform not just in the nominal case, but in the corner cases.

This project aims to make increasingly sophisticated use of machine learning and perception components in full-stack robot manipulation -- which is a bench-top surrogate for a complex autonomous system with perception in the loop. The project will involve machine learning, controls/verification, and some systems engineering to apply these ideas on real robots.
"Measuring Blood Pressure Waveforms"
Faculty Advisor: Charles Sodini
Mentor(s): Anand Chandrasehkar and Tom O'Dwyer
Contact e-mail: anandc@mit.edu
Research Area(s): Theoretical Computer Science
A doctor or nurse may have measured
your blood pressure (BP) during the last hospital visit. These BP values reflect the health of your heart
and the arteries that carry blood. While using present-day BP monitors, doctors slip your arm into a cuff
and further squeeze it to get data. This method is less accurate. As part of this project, our team is dev
eloping a device to measure blood pressure with better accuracy. The proposed device will scan the artery
with ultrasound waves and estimates blood flow velocity and area of artery to obtain the beat-to-best bloo
d pressure using basic physics equations. You may read more about this method in [A].

In this project, you will work with a team of faculty and researchers at MIT and clinicians at MGH. You w
ill learn more about electronic devices to record patient data and further explain the recorded physiologi
cal data using basic physics equations. We have divided the project into two modules:

Software Module: We have collected data from patients in the hospital using our ultrasound device. You
will help the team to analyze the data and collect more data from patients. Knowledge of signal processin
g (using MATLAB) will be helpful.

Hardware Module: We will develop flow phantom using pumps, check valves, and latex tubes. This flow ph
antom will mimic the blood flow waves in the artery. This project will help you gain hands-on experience i
n instrument design and 3D prototyping software like Solidworks.


[A]: Seo, Joohyun, et al. "Noninvasive arterial blood pressure waveform monitoring using two-element ultra
sound system." IEEE transactions on ultrasonics, ferroelectrics, and frequency control 62.4 (2015): 776-78
4.
Measuring Blood Pressure Waveforms
"Quantum networking"
Faculty Advisor: vincent chan
Mentor(s):
Contact e-mail: chan@mit.edu
Research Area(s): Communications
This project will explore the transformation of classical networking to the quantum regime. Quantum effects will greatly change the behavior of optical network algorithms and its network performance at low light levels. This exploration is a fresh look at the quantum effects of optical networking to make faint light level transactions viable. The project will first assess all the classical network layers that may be affected by quantum effects. There has been a significant amount of research in the Physical Layer on detection or low level optical signals. However, there has been little work in the next sublayer MAC (Media Access Control Layer) of the Data Link Control Layer which is the initial focus of this investigation. It is clear that familiar MAC protocols such as ALOHA and its associated “back-off algorithms” must be modified to take into account of quantum effects. Depending on the outcome this project may develop into an MEng Thesis
Strong background in probability and quantum physics (e.g., 8.05, 8.051) is required.
"Acoustic fabric microphone arrays"
Faculty Advisor: Yoel Fink
Mentor(s): Dr. Wei Yan
Contact e-mail: weiyan@mit.edu
Research Area(s): Applied Physics, BioEECS, Circuits, Inference, Materials, Devices and Photonics
Research Area(s): Functional Fibers and Fabrics, Electronics, Acoustics, AI


Audible sound is one of the important ways we communicate. Traditional technologies for sound detection, recording and acoustic communication heavily rely on rigid and cumbersome microphones, which makes their integration into fabrics challenging. In this project, we create high-performance acoustic fabric systems based on thermally-drawn acoustic fibers that are capable of detecting audible sound. The goal of the project includes 1) fabrication of thermally-drawn acoustic fibers; 2) design and fabrication of miniaturized wearable audio boards that can be directly integrated with the fabric; 3) algorithm development to allow the fabric to detect sound direction and selectively hear desired sound; 4) wireless data transmission; 5) development of AI sound recognition application
Acoustic fabric microphone arrays
"How much information can a single on-body temperature measurement derive?"
Faculty Advisor: Yoel Fink
Mentor(s): Gabriel Loke and wei Yan
Contact e-mail: gabloke@mit.edu
Research Area(s): Cognitive AI, Inference, Machine Learning, Materials, Devices and Photonics
Research Area(s): Functional Fabrics, Electronics, Healthcare, AI

10 % of the US population has diabetes, and many require constant monitoring, delivery of drugs, and regular doctor-patient meetings. In this project, we aim to design and produce a fabric which maps and monitors the body temperature variation of diabetic patients. The goals of this project are two-fold: The first is the development of a temperature-sensing fiber device. The second is the discovery of new correlations between temperature maps and diabetes using AI techniques. We are seeking SuperUROPs to assist us either on the (1) hardware or (2) software fronts. The hardware front includes fiber fabrication, post-fiber electrical and structural characterization, and set-ups for fiber demonstration. The software front includes (1) coding a digital protocol towards the aim of distributed sensing, (2) creating software visualization of the temperature map, and (3) classification of temperature maps using machine learning algorithms. The SuperUROPs working on hardware can expect to learn more about material studies on thermal drawing and electrical connections, fiber fabrication techniques, digital circuits, and sensors. The SuperUROPs working on software can expect to learn more about C++, Python, and machine learning algorithms.
How much information can a single on-body temperature measurement derive?
"Elastic functional fibers for smart textiles and neural probes"
Faculty Advisor: Yoel Fink
Mentor(s): Julieete Alain
Contact e-mail: jalain@mit.edu
Research Area(s): Applied Physics, BioEECS, Human Computer Interaction, Materials, Devices and Photonics
Faculty Advisor: Yoel Fink

Research Area(s): manufacturing, structural mechanics, polymer mechanics, neuroscience


Functional fibers, that express functionality in a single yarn, are a key step for the development of smart fabrics. Their aspect ratio also opens possibilities for multiple applications such as biological implants or neural probes. However the their appears to be a fundamental tradeoff between the conductivity of fibers and their elasticity, highly conductive fibers contain stiff metal members which limit the fibers usefulness. In this project, you will learn how to use thermal drawing to make functional fibers that have a micrometric cross-section with a highly controlled cross-sectional geometry and contain several type of materials (polymer, metals, semiconductors) while maintaining excellent elastic properties thanks to structural elasticity. This project will focus on creating elastic electrodes in an elastomeric cladding by modifying the draw process and adding mechanical steps as well as a control system to control the elongation that is built-in during the fabrication of the fiber, in order to control the final elasticity of the fiber. You will also mechanically characterize the so-obtained fiber and study the influence of the properties of the elastomeric cladding on the built-in elasticity. Further steps in the project will include incorporating these fibers in a traditional fabric weaving process and using these fibers as neural probes in the spinal cord.
A strong interest and preliminary knowledge in processing and control of mechanical processes are required as this is the core of the project. Previous experience with Matlab or LabVIEW will be preferred. Fundamentals in polymer mechanics and structural mechanics are preferred but not required.
Elastic functional fibers for smart textiles and neural probes
"A fabric that listens to your heart"
Faculty Advisor: Yoel Fink
Mentor(s): Gabriel Loke and wei Yan
Contact e-mail: gabloke@mit.edu
Research Area(s): BioEECS, Circuits, Cognitive AI, Inference, Machine Learning, Materials, Devices and Photonics
"A fabric that listens to your heart"
Research Area(s): Functional Fabrics, Electronics, Healthcare, AI


About 700,000 Americans die from heart diseases even more suffer from some form of obstructive lung disease each year, with tens of millions of people diagnosed with a heart-related problem each year. The integration of an acoustic fiber into fabrics allows for in-situ tracking of heart and lung activities. In this project, we aim to monitor and potentially perform early diagnosis of heart and lung conditions via a auscultation fabric which continuously records and monitors heartbeat signals. Our eventual goal is to be able to use advanced AI algorithms to predict and classify normal and abnormal signals to enable an on-body fabric assistant. We are seeking for UROPs to assist us in the implementation of the functional fabric towards recording signals and deciphering waveforms from both healthy and non-healthy individuals. The UROPs are expected to learn more about microphones, wearable electronics, functional fabric, healthcare, physiological monitoring, and machine learning algorithms.
A fabric that listens to your heart
"Artificially Intelligent Fabric Computers"
Faculty Advisor: Yoel Fink
Mentor(s): Gabriel Loke
Contact e-mail: gabloke@mit.edu
Research Area(s): Circuits, Cognitive AI, Inference, Machine Learning, Materials, Devices and Photonics
"Artificially Intelligent Fabric Computers"

Research Area(s): Functional Fabrics, Edge Computing, AI


Computers are trending towards form factors that are highly portable, small, personalized and even flexible. In this project we are aiming to design and fabricate computing fibers and fabrics that sense, process, store, and provide useful feedback to the wearers, hence creating a framework for intelligent clothing. We are seeking SuperUROPs to assist us either on the (1) hardware or (2) software fronts. The hardware front includes fiber fabrication, post-fiber electrical and structural characterization, and set-ups for fiber demonstration. The software front includes (1) coding in C++ for communication between microcontrollers towards the aim of distributed computing, and/or (2) coding in Python for machine learning of quantized neural networks for various Edge AI applications. Finally, we will explore the applications of these artificially intelligent fabrics in the fields of biomedicine, robotics and human augmentation. The SuperUROPs working on hardware can expect to learn more about material studies on thermal drawing and electrical connections, fiber fabrication techniques, digital circuits, computing and sensors. The SuperUROPs working on software can expect to learn more about C++, Python, machine codes, and machine learning algorithms.
Artificially Intelligent Fabric Computers
"Efficient Video Understanding for Edge Devices"
Faculty Advisor: Anantha Chandrakasan, Song Han
Mentor(s): Miaorong Wang
Contact e-mail: miaorong@mit.edu, anantha@mtl.mit.edu
Research Area(s): Circuits, Computer Graphics and Vision, Machine Learning
Video understanding is widely used in many applications, such as AR/VR, health-care and automotive. Therefore, performing video understanding with high accuracy and low energy for edge devices becomes increasingly important.
In this project, you will have a chance to:
• Work with both algorithm and hardware groups in creating high performance energy efficient video understanding solutions for edge devices
• Experiment with state-of-the-art video understanding algorithm
• Explore various neural network model compression algorithms
• Analyze the effect of model compression algorithms on hardware processing efficiency
Background in PyTorch and neural network training is helpful. Experience in FPGA is a plus.
Efficient Video Understanding for Edge Devices
"Deep Learning-Based Digital Self-Interference Cancellation for Practical Full-Duplex Radios"
Faculty Advisor: Negar Reiskarimian
Mentor(s):
Contact e-mail: negarr@mit.edu
Research Area(s): Communications, Machine Learning
Learning-based algorithms for communication systems have attracted great attention among academia and industry in the recent past. Applications of deep learning algorithms has been demonstrated in massive multiple-input multiple-output (MIMO), millimeter wave (mm-Wave) and navigation systems. More recently, a neural network-based approach for nonlinear digital SIC has been introduced that is capable of cancelling nonlinear distortion of the TX-RX leakage in full-duplex systems. However, this work does not perform well in practical scenarios. The goal of this SuperUROP project is to develop and test a deep learning-based digital SIC scheme that is able to remove both the linear and nonlinear TX self-interference. In addition, the developed algorithms should be trained on practical signals gathered from a practical wireless communication testbed equipped with full-duplex RF front-end, software defined radios (USRPs) and GNU radio software platform.
Desired Skills:
RF communications, cellular waveforms, and signal processing concepts.
System-level understanding of wireless RF communication systems, cellular waveforms, and signal processing concepts.
Prior experience with USRPs and GNU radio is highly desired.
Experience with programming languages such as Python, C/C++ is strongly desired.
Deep Learning-Based Digital Self-Interference Cancellation for Practical Full-Duplex Radios
"TinyML and Efficient Deep Learning on Edge Devices"
Faculty Advisor: Song Han
Mentor(s): none
Contact e-mail: songhan@mit.edu
Research Area(s): Computer Architecture, Computer Graphics and Vision, Computer Systems, Machine Learning
(1) TinyML and efficient inference on micro-controllers: model compression; binary/ternary networks; neural architecture search.

(2) hardware accelerator for efficient video recognition (Temporal Shift Module): Link

(3) sparse accelerator; deep learning accelerators with low latency and low power;

- design Tiny ML models
- writing fast kernels with SIMD instructions (if you've taken 6.172 that would be great)
- system architecture design
- RTL, FPGA development
- substantial time commitment is required

more details available at:
Link
Link
"Implementing a Faster Fast Gauss Transform"
Faculty Advisor: Pablo A. Parrilo
Mentor(s): Jason Altschuler
Contact e-mail: parrilo@mit.edu
Research Area(s): Machine Learning, Numerical Methods, Signals and Systems
The Discrete Gauss Transform is the computational bottleneck task in many important problems in machine learning, scientific computing, and applied mathematics. In this SuperUROP, you will learn about and implement a recently developed state-of-the-art algorithm for the Discrete Gauss Transform. The end goal is to develop production-quality software that can be posted open-source online and be easily used by researchers around the world, as well as to investigate possible extensions.

In addition to mentoring from the PI, you will also have access to a PhD student directly supervising and mentoring you.

Prerequisites:
CS: Strong coding background (e.g., Python, Julia, etc). Competitive programming experience is not required, but may be helpful.
Math: Basic proficiency with linear algebra and multivariate calculus. Interest in learning about spherical harmonics, although no previous background necessary.

Contact: Please contact Prof. Pablo Parrilo (parrilo@mit.edu) if you are interested in working on this project.
Implementing a Faster Fast Gauss Transform
"Linguistic Analysis of Wikipedia for Question Answering"
Faculty Advisor: Boris Katz
Mentor(s): Sue Felshin
Contact e-mail: boris@mit.edu
Research Area(s): Cognitive AI, Human Computer Interaction, Natural Language and Speech Processing
We study how natural language processing and information access methods can improve human-computer interaction. Wikipedia, the world's largest crowdsourced encyclopedia, is a great source of knowledge for question answering systems. Using linguistic analysis of information in Wikipedia and related resources, we hope to considerably increase the precision and coverage of our START Natural Language Question Answering System. The project will involve our component systems that turn Wikipedia infoboxes and other structured and semi-structured data into a virtual database accessible via natural language and extend START's parsing and question answering capabilities. Linguistic Analysis of Wikipedia for Question Answering
"Understanding the limits of machine and human vision"
Faculty Advisor: Boris Katz
Mentor(s): Andrei Barbu
Contact e-mail: boris@mit.edu
Research Area(s): Cognitive AI, Computer Graphics and Vision, Machine Learning
What are the limits of machine vision? Why is it that human vision is so reliable? Help us understand the answers to both of these questions. Over the past few years, our group has developed ObjectNet, a new kind of challenging vision benchmark. It is challenging because it contains controls: the images are collected in a systematic way varying properties of images such as their backgrounds and object rotations. We know that machines perform poorly on ObjectNet, but why is that? Is something missing from current networks? This project involves both building better detectors and gathering human data to pin down answers to these questions. Understanding the limits of machine and human vision
"Leveraging machine learning to transform climate modeling"
Faculty Advisor: Raffaele Ferrari
Mentor(s): Andre Souza and Ali Ramadhan
Contact e-mail: raffaele@mit.edu
Research Area(s):
We invite students to join our group to develop, train, and test new models that better represent the ubiquitous turbulence that characterizes flows in the global ocean. You would participate in the Climate Modeling Alliance (clima.caltech.edu; CilMA), a coalition of scientists, engineers, and applied mathematicians from Caltech, MIT, the Naval Postgraduate School, and NASA’s Jet Propulsion Laboratory building a new climate model using the Julia programming language that leverages recent advances in the computational and data sciences to learn directly from a wealth of Earth observations from space, ocean and land to improve the fidelity of climate model projections.

CliMA has been established because climate models cannot represent every cloud in the atmosphere and every wave in the ocean. Instead they rely on ad-hoc "parameterizations" or surrogate models to represent the effect of physics at scales smaller than grid-scale of the climate model. Existing parameterizations are often quite inaccurate and largely contribute to the uncertainty in projections of future climate change. CliMA aims to improve the fidelity of climate models by developing “parameterizations” trained against observations and very high resolution turbulence simulations like the ones shown in the figure.

Potential projects include using machine learning methods such as neural differential equations and Gaussian process regression to develop new parameterizations of turbulence in the upper ocean from high-resolution simulations. Upper ocean turbulence is key in climate projections, because it sets the rate of heat and carbon ocean uptake among other climatically important processes. We are also interested in performing hyperparameter optimization, model selection, and Bayesian parameter inference using probabilistic programming to fully quantify the performance of each data-driven parameterization.

There are plenty of open problems and we're just getting started so there are lots of opportunities to try different things and make a fundamental contribution to the field!
Leveraging machine learning to transform climate modeling
"Exploring Security Vulnerabilities in Modern Processors"
Faculty Advisor: Mengjia Yan
Mentor(s):
Contact e-mail: mengjia@csail.mit.edu
Research Area(s): Computer Systems
Modern processors have been aggressively optimized for performance and energy efficiency. However, recent attacks, such as high-profile Spectre and Meltdown attacks, have shown how vulnerable modern computer hardware is.

This projects is to explore various side channel vulnerabilities in modern processors. We are going to study potential security problems in those micro-architecture structures, which have not been fully understood by the community, such as buffers and directories in the memory hierarchy and network-on-chip.

The project will involve reverse engineering commercial processors, developing end-to-end attacks and breaking privacy of modern applications, including crypto libraries and ML applications.

References: Link Link

Prerequisites:
- Performed well in 6.004
- Taking 6.823 is a plus
Exploring Security Vulnerabilities in Modern Processors
"Simulating artificial spiking neural networks made from superconducting nanowires"
Faculty Advisor: Karl Berggren and Nancy Lynch
Mentor(s):
Contact e-mail: berggren@mit.edu
Research Area(s): Applied Physics, Circuits, Machine Learning, Materials, Devices and Photonics, Nanotechnology, Theoretical Computer Science
Devices that naturally mimic the dynamics of the human brain are actively being developed for use in spiking neural networks, with the ultimate goal of achieving fast, energy-efficient computation. Recently, an artificial neuron made from superconducting nanowires has shown promise as a neuromorphic device with competitive energy performance(1). This project will build on the previous nanowire neuron design by using simulations to study how the neuron can be expanded into larger networks and solve problems. The student will demonstrate the neuron’s scalability by investigating how common issues like fabrication variances and noise impact the device’s operating margins. The student will also use a combination of algorithms and simulations to create large circuits designed for applications that are uniquely suited to spiking neural networks, such as pattern recognition and understanding biological dynamics. Experience with LTSpice, circuit design, and/or algorithms is preferred.

1.Toomey, E., Segall, K. & Berggren, K. K. Design of a Power Efficient Artificial Neuron Using Superconducting Nanowires. Front. Neurosci. 13, (2019).
Simulating artificial spiking neural networks made from superconducting nanowires
"Constellation"
Faculty Advisor: Max Goldman
Mentor(s): none
Contact e-mail: maxg@mit.edu
Research Area(s): Human Computer Interaction, Programming Languages (incl software eng)
Constellation enables collaborative programming in the Eclipse IDE -- think Google Docs for Eclipse. It was designed for active learning in 6.031 Software Construction, where students do pair programming exercises nearly every class meeting using Constellation. In addition to collaborative editing, the system also allows course staff to review and assess student work.

See: Link

Some possible research projects are below. On any project, you will review related work; design and implement new capabilities; evaluate them either in deployment to 6.031 or in a lab study; and report on the results.

- Improved collaborative editing: add missing collaboration features that help students focus on learning rather than logistics
- Visualizing pairs: better visualizations that capture more of each pair's process and results, or that help staff identify struggling students and gauge the progress of an entire multi-hundred-student class
- Staff feedback: enable course staff to give high-quality feedback more quickly, and deliver their comments effectively to students
- Beyond the classroom: e.g. Constellation could enable on-line office hours where staff help remotely

Prerequisites: 6.031, experience with web development (e.g. 6.170) and HCI
"Engineering transcriptional regulators to control immune functions of anti-tumor T cells"
Faculty Advisor: Timothy Lu
Mentor(s): Tim (TingXi) Guo
Contact e-mail: timlu@mit.edu, tingxi@mit.edu
Research Area(s): BioEECS
Primary T lymphocytes can be genetically engineered ex vivo to produce effective anti-tumor cell therapies. To improve the safety and efficacy of this approach, our project aims to develop the next generation of therapeutic T cells, using synthetic promoters to precisely regulate effector responses. We recently generated a library of synthetic promoters for discovering cell state-specific transcriptional response elements in a high-throughput fashion. Using this platform in human T cells, we will identify inducible promoters that are activated by tumor-associated signals. Disease-responsive promoters can be used to express a variety of user-defined therapeutic programs. Students interested cancer immunology, cell therapies, and synthetic biology are strongly encouraged to apply. Familiarity with molecular biology and/or basic wet-lab techniques is preferred. Engineering transcriptional regulators to control immune functions of anti-tumor T cells
"Prediction of Clinical Outcomes from Electrocardiograms (ECGs) using Machine Learning"
Faculty Advisor: Collin Stultz
Mentor(s):
Contact e-mail: cmstultz@mit.edu
Research Area(s): Machine Learning
The electrocardiogram (ECG) provides a wealth of information about the health of the cardiovascular system and the computational cardiovascular research group (CCRG) has a longstanding interest in applying machine learning methods to predict cardiovascular events; e.g., death, heart attack, stroke. The primary goal of this collaborative project is therefore to develop deep learning models that use the ECG to predict adverse outcomes. The SuperUrop would have appointments at MIT and at the Massachusetts General Hospital (MGH), and work closely with MGH physicians to help develop these models.

Prerequisites
• Must be able to commit at least 10 hours per week
• Proficiency with Python
• Knowledge of deep learning and/or machine learning (6.036, 6.862, or experience with building/applying deep neural networks)
• Interest in working at the intersection of medicine and machine learning
Prediction of Clinical Outcomes from Electrocardiograms (ECGs) using Machine Learning
"Accelerating materials discovery with artificial intelligence"
Faculty Advisor: Professor Heather J. Kulik
Mentor(s): Mr. Aditya Nandy
Contact e-mail: hjkulik@mit.edu
Research Area(s):
Although the discovery and synthesis of new materials, catalysts, and functional molecules represents the foremost effort that unifies thousands of researchers, presently characterized compounds represent a minute fraction (ca. 1 part in 10^50) of
chemical space. First-principles, physics-based computation has emerged as a valuable tool to rationalize origins of emergent phenomena in new materials, even guiding materials synthesis. However, both accelerated physics-based computation and experiment are too slow to navigate this space to discover new materials. Although machine learning models can be trained on physics-based models to produce results that would normally take hours or weeks in less than a second, replacing experimental observations remains a key challenge. This project will focus on the extraction of experimental observations from the literature using a combination of sentiment and graph analysis on metal-organic framework and other inorganic materials known for their desirable catalytic, electronic, and magnetic properties. The project will involve the development of machine learning models and the testing of these models to guide materials design efforts in the Kulik group ( Link ). This broader project effort in the Kulik group involves development of databases, a web interface, and will integrate with our open source code molSimplify and its automatic design tools. The UROP will become an integral part of our interdisciplinary physics-based and development-based materials discovery team.
Accelerating materials discovery with artificial intelligence
"High-level processing module for a speech analysis system"
Faculty Advisor: Dr. Stephanie Shattuck-Hufnagel
Mentor(s):
Contact e-mail: sshuf@mit.edu
Research Area(s): Natural Language and Speech Processing
Current automatic speech recognition systems function usefully, but operate very differently from human speech perception. This UROP involves work on a speech signal analysis system that is modeled more closely on what we know about human speech processing. The project involves developing a consolidator module to integrate acoustic, lexical, and prosodic information derived from the signal into a preliminary hierarchical structure for the entire phrase or utterance, even before the speaker’s intended words are fully recognized. Candidates with overlapping interests in Course 6, Course 9 and/or Course 24 are particularly appropriate. Contact Dr. Stefanie Shattuck-Hufnagel, Speech Communication Group, sshuf@mit.edu. High-level processing module for a speech analysis system
"Acoustic phonetic variation in speech recognition"
Faculty Advisor: Stefanie Shattuck-Hufnagel
Mentor(s): none
Contact e-mail: sshuf@mit.edu
Research Area(s): Natural Language and Speech Processing
Words are pronounced very differently in different contexts, and it is challenging to find a vocabulary to describe these differences in enough detail to be useful both for recognizing the words in an automatic speech recognition system and understanding the principles that govern these patterns of variation. This project involves creating an inventory of systematic context-dependent surface variation, determining the factors that govern it, and evaluating hypotheses about the principles that underlie this phenomenon, including massive reductions like "Why don't you" (approximately) wynchah, or I'm going to (approximately) ahmuhnuh. Candidates with overlapping interests in Course 6, Course 9 and/or Course 24 are particularly appropriate. Contact Dr. Stefanie Shattuck-Hufnagel, Speech Communication Group, sshuf@mit.edu
"Website for a speech analysis toolkit"
Faculty Advisor: Stefanie Shattuck-Hufnagel
Mentor(s): none
Contact e-mail: sshuf@mit.edu
Research Area(s): Natural Language and Speech Processing
This project involves completing a semi-interactive website for an ongoing project on the modeling of human speech perception, with implications for automatic speech recognition systems. The website will include an existing tutorial on labelling the linguistically significant information in the speech signal, labelling tools for displaying and annotating the signal and analyzing the results, and several labeled speech databases, along with a prototype speech analysis system and relevant publications and theses. Contact Dr. Stefanie Shattuck-Hufnagel, Speech Communication Group, sshuf@mit.edu
"Brain Algorithms"
Faculty Advisor: Nancy Lynch
Mentor(s):
Contact e-mail: lynch@csail.mit.edu
Research Area(s): BioEECS, Cognitive AI, Computational Biology, Machine Learning, Theoretical Computer Science
We have been working on an algorithmic theory to try to describe how the brain solves a variety of problems involving focus and attention [1], neural representation of concepts [2], and learning [3]. All our work is based on a synchronous Spiking Neural Network (SNN) model. Our approach is mostly theoretical---defining models, designing and analyzing algorithms, and proving lower bounds---although we are motivated by the hope of understanding behaviors of actual brains.

In this project, you will simulate some of the algorithms that we have developed, to see whether their behavior in practice matches what is predicted by the analysis, or is even better. In particular, we are interested in simulations of biologically plausible, local learning algorithms for hierarchically-structured data like those in [3], plus some of proposed extensions (to different types of data hierarchies, variations in the network model, and component failures). One possibility is to compare our biologically-plausible learning algorithms with standard centralized gradient-descent Machine Learning algorithms, in terms of both efficiency of learning, and the types of representations produced. If the representations are similar, we might be able to provide some explanation for the success of the ML algorithms. If they are different, that would also be interesting. The project should also provide opportunities for algorithm design and theoretical analysis.

Ideal background: Programming with Python, numerical simulation, probabilistic modeling and analysis, design and analysis of algorithms, machine learning, interest in brain behavior.

References:
[1] Nancy Lynch, Cameron Musco, and Merav Parter. Winner-Take-All Computation in Spiking Neural Networks. ArXiv:1904.12591, April 2019.

[2] Yael Hitron, Nancy Lynch, Cameron Musco, and Merav Parter. Random Sketching, Clustering, and Short-Term Memory in Spiking Neural Networks. 11th Symposium on Innovations in Theoretical Computer Science (ITCS 2020). January 2020.

[3] Nancy Lynch and Frederik Mallmann-Trenn. Learning of Hierarchically Structured Concepts. ArXiv:1909.04559, February 2020.
Brain Algorithms
"Dynamical aspects of retinal neural coding for a moving object"
Faculty Advisor: Nancy Lynch
Mentor(s): Brabeeba Wang
Contact e-mail: brabeeba@mit.edu
Research Area(s): BioEECS, Cognitive AI, Computational Biology, Theoretical Computer Science
A major question in visual neuroscience is what visual information is represented in the retinal neural coding in retina ganglion cells. In recent decades, neuroscientists have gained a detailed understanding of different cell types and microcircuit signatures in the retina. An emerging picture is that the retina does not encode merely static images, but also dynamical features such as speed and direction of motion; even extracting static features like contours involves dynamic input, generated as a result of eye movement. However, so far, there is little understanding of the nature of retinal neural coding in the dynamical context.

In this SuperUROP project, you will simulate detailed biological circuits following biological details from a line of work by Prof. Markus Meister [1]. The goals are to understand (1) the principle of retinal neural coding of movement of objects, (2) the role of Short-Term Potentiation and Depression (STP/STD) in the dynamical adaptation of the retinal neural code, and (3) plasticity in the retina during development.

Recommended background: You should be comfortable with numerical simulation of differential equations and with reading neuroscience papers, know basic probability, and be enthusiastic about understanding how the brain processes visual information.

[1] Eye smarter than scientists believed: Neural computations in circuits of the retina. Gollisch, T., and Meister, M. Neuron 65:150-164 (2010).
Dynamical aspects of retinal neural coding for a moving object
"Efficient Computing for Robotics on an FPGA"
Faculty Advisor: Vivienne Sze & Sertac Karaman
Mentor(s): none
Contact e-mail: sze@mit.edu
Research Area(s): Circuits, Computer Architecture, Robotics, Signals and Systems
Autonomous navigation of miniaturized robots (e.g., nano/pico aerial vehicles) is currently a grand challenge for robotics research, due to the need for processing a large amount of sensor data (e.g., camera frames) with limited on-board computational resources. Enabling efficient computing is critical for this task. Our group has been investigating various approaches from both the algorithm and hardware perspective, including both deep neural network (DNN) and non-DNN based solutions.

Example projects include
Navion: Link
FastDepth: Link
Robot Exploration: Link

The goal of this project is to contribute to the development of an energy-efficient FPGA platform, and potentially efficient algorithms, for various forms of processing to support real-time navigation. There are several opportunities to get involved in this project ranging from:
** RTL design for the FPGA
** Development embedded software for an ARM core
** Development of efficient DNN and non-DNN algorithms

Prior experience in FPGA design and/or algorithm design is a plus.
Efficient Computing for Robotics on an FPGA
"Robust, interpretable machine learning"
Faculty Advisor: Tommi Jaakkola
Mentor(s): Guang-He Lee
Contact e-mail: guanghe@mit.edu
Research Area(s): Machine Learning
Modern AI techniques invariably rely on highly flexible black-box models such as deep neural networks. While effective in the predictive sense, the models often reveal little about their underlying mechanism of reasoning responsible for the predictions. Lack of transparency is particularly limiting in medical, legal, security, or financial applications where decisions need to be vetted and confirmed prior to acting on them. Broadly, our goal is to develop reliable machine learning models that produce robust, interpretable predictions. We are interested in solutions grounded in theory yet effective for practical real-world deployment. As part of this project, students may explore extensions to prior approaches to interpretability, robustness, or develop new approaches building on invariant prediction. Robust, interpretable machine learning
"Autonomous Driving"
Faculty Advisor: Daniela Rus
Mentor(s): none
Contact e-mail: rus@csail.mit.edu
Research Area(s): Cognitive AI, Control and Decision Systems
In this project we are developing perception, planning, control, and machine learning algorithms for autonomous driving. We are emphasizing the capabilities of autonomous vehicles in difficult driving situations such as congestion or weather, and the interactions between autonomous vehicles and human-driven vehicles. We are aiming to develop new algorithms that increase the range of capabilities of the robots and that have provable guarantees.
"Printable Robots"
Faculty Advisor: Daniela Rus
Mentor(s): none
Contact e-mail: rus@csail.mit.edu
Research Area(s): Cognitive AI, Human Computer Interaction, Materials, Devices and Photonics
Designing and fabricating new robotic systems is typically limited to experts, requiring engineering background, expensive tools, and considerable time. In contrast, to facilitate everyday users in developing custom robots for personal use, this project aims to create algorithms, systems, and tools to easily create printable robots from high-level structural specifications. From that, the system generates complete mechanical drawings suitable for fabrication, instructions for the assembly of electronics, and software to control and drive the final robot.This project aims to develop steps towards creating a hardware compiler.
"Soft Robots"
Faculty Advisor: Daniela Rus
Mentor(s): none
Contact e-mail: rus@csail.mit.edu
Research Area(s): Cognitive AI, Materials, Devices and Photonics, Numerical Methods, Theoretical Computer Science
In this project we address the design, fabrication, and control of a soft robots such as robot fish and robot manipulators. Soft robot generally consist of several segments actuated using bidirectional fluidic elastomer actuators. The robots are fabricated using molding and 3d printing processes. We also develop the associated computation and control systems enable the robots to move.
"AI is for Kids"
Faculty Advisor: Hal Abelson
Mentor(s): none
Contact e-mail: hal@mit.edu
Research Area(s):
The MIT App Inventor group (appinventor.mit.edu) works to empower young people worldwide to create original applications with mobile technology. This work will extend the App Inventor platform to encompass the neural network features in the emerging generation of smartphones and create new project opportunities for kids to build AI applications.

Interested students should see the project Web site at
appinventor.mit.edu
"Can Neural Networks Be Trusted?"
Faculty Advisor: Luca Daniel
Mentor(s): Tsui-Wei (Lily) Weng
Contact e-mail: luca@mit.edu
Research Area(s): Numerical Methods
As artificial intelligence (AI) systems automate more tasks, the need to quantify their vulnerability and alert the public to possible failures has taken on new urgency, especially in safety-critical applications like self-driving cars and fairness-critical applications like hiring and lending. To address the problem, this project aims at developing methods that efficiently quantify how much each individual input (e.g. an image of an ostrich) can be altered before the neural network misclassifies it (e.g. classifying the ostrich as a vacuum cleaner!), on its own or through a malicious attack. Here is an example of a task within this project developed by Akhilan Boopath, a superurop student in our group: Link . Future tasks may include expanding our existent framework to larger, and more general neural networks, and developing tools to quantify the level of vulnerability based on many different ways of measuring input-alteration. Can Neural Networks Be Trusted?
"Information Sharing among Distributed Neural Networks"
Faculty Advisor: Dorothy Curtis
Mentor(s): Aviv Segev
Contact e-mail: dcurtis@csail.mit.edu
Research Area(s): Computer Systems
Neural networks and especially the deep learning approach have displayed successful results in areas such as image processing, medicine, and the stock market. Neural networks perform well when they are built for a specific task and the set of inputs and set of outputs are well defined. However, these accomplishments are very limited in scope, and communication between different neural networks to share knowledge that will lead to the performance of more general tasks is still inadequate.
Communication between (neural network) specialists is the goal of this project. We propose to utilize independent sets of neural networks trained for specific tasks, while allowing knowledge transfer between each of the neural networks. The idea is based on computer network architecture, which is a communication system that transfers data between distributed computers. The idea will similarly allow each neural network to specialize in its own task while transferring and receiving information from other neural networks. This will allow different neural networks to be plugged in as an open platform. It will also allow additional information to be requested, when the task at hand is difficult or hard to resolve.
The project will try to bridge between: knowledge transfer; neural network communication; neural network Internet; distributed systems; and communication between specialists.
Information Sharing among Distributed Neural Networks

Total: 135

Technical questions related to this site? Please contact Search-Master@superurop-apply.mit.edu