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

SuperUROP Projects

"Robust Perception (for Robots)"
Faculty Advisor: Russ Tedrake
Mentor(s):
Contact e-mail: russt@mit.edu
Research Area(s): Artificial Intelligence, Control, Signals and Systems
Autonomous cars will soon be driving my kids to school. Robots will be moving through are homes picking things up. And they will be using perception algorithms based on deep learning. Perception algorithms that work really well a lot of the time, but we know they also make mistakes. That's terrifying!

The Robot Locomotion Group at CSAIL is trying to understand the problem of how to formalize concepts of robustness for robots/ feedback systems with complex perception systems in the loop. Understanding where these systems makes mistakes is a good start. Proving that they don't make horrible mistakes is better. Generating fundamentally new perception systems that give stronger robustness guarantees is the home run.

We are looking for SuperUROPs to help with this process, which could include experiments on our robot hardware (up to an including a 400 lb humanoid) to investigating novel optimization/verification algorithms.
Robust Perception (for Robots)
"Robust Manipulation"
Faculty Advisor: Russ Tedrake
Mentor(s):
Contact e-mail: russt@mit.edu
Research Area(s): Artificial Intelligence, Control
Despite the amazing youtube videos, the truth is that most robots still fail to pick up most objects, most of the time. Manipulation in unstructured environments will be one of the key technology break-throughs in robotics over the next few years. The Robot Locomotion Group in CSAIL is attempted to address this problem by understanding the mathematical structure of the decision making problem using combinatorial and continuous (ideally convex) optimization. To make these optimizations tractable, we need to leverage the structure in the equations of motion of the mechanical system.

We are looking for SuperUROPs to help with this process, which could include experiments on our robot hardware (up to an including a 400 lb humanoid), to investigating optimization algorithms, to focusing on potential insights from mechanics/control.
Robust Manipulation
"Hexahedral Meshing Benchmarks"
Faculty Advisor: Justin Solomon
Mentor(s): Paul Zhang
Contact e-mail: pzpzpzp1@gmail.com
Research Area(s): Applied Physics, Graphics and Human-Computer Interfaces, Numerical Methods, Theoretical Computer Science
Hexahedral meshing (hex meshing) has advantages over tetrahedral meshing for finite element simulation in both accuracy and the number of mesh elements required. With the recent increase in hex meshing techniques, it is unclear which methods generate better hex meshes. This project will be to determine a hex meshing benchmark over a variety of meshes to evaluate hex meshing techniques. Another product of this project will be a database of hex meshes with varying structure and topology.

Useful skills:
C++, or Matlab experience
Enthusiasm for simulation and meshing
Hexahedral Meshing Benchmarks
"Machine Learning for Medicine"
Faculty Advisor: Roger Mark
Mentor(s): Li-Wei Lehman
Contact e-mail: lilehman@mit.edu
Research Area(s): BioEECS
Project Description: The Laboratory for Computational Physiology at MIT’s Institute of Medical Engineering and Science (IMES) is seeking a highly-motivated student to participate in research in Machine Learning for Medicine. Successful candidate will join a multidisciplinary research team to apply machine learning and statistical techniques for patient risk stratification and treatment decision support in critical care.

Prerequisite: The ideal candidate would have taken courses in advanced machine learning, statistical inference, or related courses. Knowledge and experience in one or more of the following areas would be desirable: deep learning, reinforcement learning, signal processing. Familiarity with Python, MatLab, or R preferred.

Apply: In addition to a resume, applicants should specify their research interests in their email to Dr. Li-wei Lehman
"Exploring new frameworks and paradigms for implementing reliable distributed and IoT systems"
Faculty Advisor: Armando Solar-Lezama
Mentor(s): Ivan Kuraj
Contact e-mail: ivanko@csail.mit.edu
Research Area(s): Computer Systems
Many existing frameworks let developers write distributed systems by abstracting low-level details of the underlying aspects such as communication, protocols, consistency, and failures. We’re working on a framework that takes these ideas further, allowing the programmers to specify their intended distributed computation in more intuitive
ways, while letting a smart compiler to emit the final optimized distributed program that runs across multiple nodes in the system. The translation exploits some existing technologies and techniques for final implementations such as the actor concurrency model (Akka), distributed protocols and algorithms, and processing systems (Spark), with plans of enlarging the set for IoT and edge computing support. The project lies at the intersection of distributed systems, distributed algorithms, and programming languages.
We are looking for a prospective student to contribute to one of two main efforts: 1) extending the compiler support for lower-level IoT/embdedded devices (e.g. Arduino) and protocols (e.g. ZigBee); 2) development of a cutting-edge run-time system for communication between machines/devices, both inside and outside "the cloud". The focus of this project is for students to learn about modern distributed programming techniques and frameworks, and gain experience in development and benchmarking of state-of-the-art IoT, big data, and potentially embedded systems.
"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
"THz Image Sensors with Electronic Beam Forming"
Faculty Advisor: Ruonan Han
Mentor(s):
Contact e-mail: ruonan@mit.edu
Research Area(s): Circuits, Signals and Systems
Terahertz wave is a safe (compared to X-ray) and high-resolution (compared to microwave) tool for non-invasive imaging in security and industrial quality control. We have developed focal-plane detector arrays for THz imaging. However, due to the large wavelength (~0.5mm) that determines the on-chip antenna size, it is impossible to integrate a large number of THz pixels on a single chip. Because of this, THz imagers often use mechanical scanner which is bulky and slow. In the next generation of THz image sensors, the pixels detect not only the intensity of the incident THz wave, but also its phase. It is then possible to achieve electronic scanning inside a 2D space; hence the mechanical scanner is eliminated. The SuperUROP in this project will have the opportunity to (1) build the test board for beam steering and data sampling, (2) develop signal processing algorithms that reconstruct the images from the raw data of our chips. THz Image Sensors with Electronic Beam Forming
"Design of a Piezoelectric (PZ) Energy Harvesting Device (EHD) for Wide-Bandwidth Harvesting of Vibration Energy"
Faculty Advisor: Jeff Lang
Mentor(s):
Contact e-mail: lang@mit.edu
Research Area(s): Energy
Harvesting energy from the environment is a key technology that is being developed to power sensor nodes in the Internet of Things (IoT). This project focuses on harvesting vibrational energy using a PZ EHD. When vibration energy excites such a device at its mechanical resonance frequency, adequate output power is achieved. But, as shown to the right, for excitations away from resonance, the power decreases rapidly. We have developed an electrical loading technique, refered to as “Bias Flip”, to “tune”, or shift, the resonance frequency of the EHD. This has proven to be successful for small frequency mismatches, but we want to achieve high output power over an even wider bandwidth. To achieve this, we will design a new PZ EHD for optimum bandwidth when used in conjunction with Bias Flip circuitry. This project will involve two phases.
1) Optimization of the EHD structure using simulation. We will contact an external company to produce the new EHD to our specification.
2) Setting up a Bias-flip test bench and testing the optimized EHD.
We plan to publish the use of BF with new PZ EHDs in the spring of 2019. This work will hopefully be included in the publication.
Design of a Piezoelectric (PZ) Energy Harvesting Device (EHD) for Wide-Bandwidth Harvesting of Vibration Energy
"Crowdsourcing attention and interestingness of images"
Faculty Advisor: Aude Oliva
Mentor(s): Zoya Bylinskii
Contact e-mail: zoya@mit.edu
Research Area(s): Artificial Intelligence, Graphics and Human-Computer Interfaces
Where do people look in images, what do they find interesting in graphic designs, and how do they explore 2D and 3D scenes? This project aims to capture human attention on different types of images through interactive crowdsourcing tasks, on the web and on mobile. Aspects of this project include building user interfaces for large-scale crowdsourcing experiments, statistical data analysis, and machine learning to build predictive models of attention and visual exploration. Applications include smart image compression, automatic thumbnailing, and personalized graphic designs. This project lies at the interface of human vision, computer vision, and HCI. Crowdsourcing attention and interestingness of images
"GaN Power Electronics"
Faculty Advisor: Tomas Palacios
Mentor(s):
Contact e-mail: tpalacios@mit.edu
Research Area(s): Materials and Devices
High-power transistors fabricated with gallium nitride (GaN) are revolutionizing power electronics in applications such as electrical vehicles, data centers, smart grids, etc.
Despite of the exciting results demonstrated so far, GaN power devices are still far from their theoretical limits. One area that is not yet fully understood is the off-state breakdown mechanisms which limit the high voltage operation of devices. This project aims to provide fundamental insight on the breakdown mechanism in lateral GaN High-Electron-Mobility Transistors (HEMTs) by characterizing reverse-biased devices under illumination and over a wide range of temperatures. This will enable us to develop more accurate models for breakdown in GaN and to push the operating limits of these promising devices.
GaN Power Electronics
"Privacy and machine learning"
Faculty Advisor: Lalana Kagal
Mentor(s):
Contact e-mail: lkagal@csail.mit.edu
Research Area(s): Artificial Intelligence, Computer Systems
Enterprises usually provide strong controls to prevent cyberattacks and inadvertent leakage of data to external entities. However, when it is necessary to analyze and derive insights from the data, there are insufficient controls and employees are usually permitted access to all information about customers including sensitive and private information. Though it is important to be able to identify useful patterns of one's customers for better customization and service, customers' privacy must not be sacrificed to do so. We propose an alternative - a framework that will allow analytics over big data while preserving privacy. We are developing an efficient and scalable differential privacy framework for Apache Spark that provides strong privacy guarantees for users even in the presence of an informed adversary, while still providing high utility for analysts. We are looking to evaluate this work with big data from different domains such as healthcare and financial, as well as extend the functionality of the system.

Familiarity with Spark and Scala will be helpful.
Privacy and machine learning
"Semantic Smart Contracts for Enforceable Data Sharing Agreements"
Faculty Advisor: Lalana Kagal
Mentor(s):
Contact e-mail: lkagal@csail.mit.edu
Research Area(s): Computer Systems
A smart contract is a piece of code that is stored on a blockchain, triggered by blockchain transactions and that reads and writes data in that blockchain's database. We would like to develop a high level language for specifying data sharing contracts that bridges the gap between programming languages (are currently used for smart contracts) and natural language (currently used for legal agreements). These contracts will focus on enabling the sharing of sensitive information such as healthcare information. The contracts will describe responsibilities of both the subject and consumer of the data, tie parties to their duties, ensure payment, when applicable and hold violators accountable.

Familiarity with Ethereum and Solidity is a plus but not required.
Semantic Smart Contracts for Enforceable Data Sharing Agreements
"A Decentralized Architecture for the Web"
Faculty Advisor: Lalana Kagal
Mentor(s):
Contact e-mail: lkagal@csail.mit.edu
Research Area(s): Computer Systems, Graphics and Human-Computer Interfaces
Social networking forms an important part of online activities of Web users. Web sites such as Facebook have millions of users using them everyday. However, these sites present two problems. Firstly, these sites form information silos. Information on one site is not usable in the others. Secondly such sites do not allow users much control over how their personal information is disseminated, which results in potential privacy problems. We are developing an open source project that aims to radically change the way Web applications work today, resulting in true data ownership as well as improved privacy. We are developing protocols that allow user data to be managed separately from applications, resulting in decentralized social software. Users will be free to switch systems while keeping the same social connections, and developers will be free to innovate without having to worry about capturing a user base or maintaining a massive back-end operation. We believe the result will be a dramatic increase in innovation with an increased respect for user autonomy and privacy.

Experience with JavaScript Web application development is a requirement, and familiarity with Linked Data is a plus.
A Decentralized Architecture for the Web
"Real-time Planning Support for ICU Work Rounds"
Faculty Advisor: Roger Mark
Mentor(s): Alistair Johnson
Contact e-mail: aewj@mit.edu
Research Area(s): Artificial Intelligence, BioEECS, Graphics and Human-Computer Interfaces
Intensive care units (ICUs) provide lifesaving treatment for critically ill patients. Within the ICU healthcare providers meet routinely, usually once daily, for a scheduled discussion where the clinical information for a patient is reviewed and a care plan is developed. These discussions, referred to as “rounds”, allow individual providers to contribute to the overall care plan based upon their specific expertise and their familiarity with the patient. The ICU is a data rich environment and one could view rounds as a discussion where each care provider summarizes their perspective on the patient’s data which they have reviewed. Our hypothesis is that predictive algorithms can provide a new fully data-driven perspective of the patient and that this perspective could improve communication about patient status, inform the development of care plans, and improve patient outcomes. This project will involve developing predictive models for specific treatment strategies and displaying the results of the models in an interpretable fashion during rounds. We will discuss the most impactful application for the predictive models with senior nurse practitioners at the Beth Israel Deaconess Medical Center (BIDMC), and the project will leverage a real-time stream of patient data currently established between the BIDMC and MIT. Real-time Planning Support for ICU Work Rounds
"Diffraction ray tracing"
Faculty Advisor: Fredo Durand
Mentor(s):
Contact e-mail: fredo@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
The goal of this project is to simulate wave effects (diffraction, interference) using a path tracing approach, by tracing additional rays corresponding to fake scattering events.
"AI with Computer Graphics"
Faculty Advisor: Fredo Durand
Mentor(s): fredo@mit.edu
Contact e-mail: fredo@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Our goal is to use 3D graphics and create virtual worlds to train and test AI and machine learning algorithms. We build methods to procedurally create realistic scenes of cities and natural environments. Super-urop opportunities include:
- Hard negative mining. Use gradient descent, metropolis sampling (MCMC) and other strategies to find a plausible 3D scene that makes the job of a neural network maximally hard.
- Sobolev training: force neural network output to be invariant to phenomena such as lighting and pose by leveraging derivatives of training data wrt these effects.
AI with Computer Graphics
"3D graphics for AI"
Faculty Advisor: Fredo Durand
Mentor(s):
Contact e-mail: fredo@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Our goal is to use 3D graphics and create virtual worlds to train and test AI and machine learning algorithms. We build methods to procedurally create realistic scenes of cities and natural environments. Super-urop opportunities include:
- the generation of weathered or dirty looks. 3D renderings often look too clean. Computer graphics techniques that make objects look dirty tend to be expensive. We want to develop techniques to make things such as buildings, cars, roads, more realistic and dirty. This can be achieved using machine learning or direct procedural approaches,
- Weather simulation: Simulate rain, snow and fog. Simulate dynamic clouds.
- Night rendering. Simulate urban scenes at night. One challenge is scalability with respect to the number of light sources.
- Automatically generate building and car interiors.
- Road mirages. Simulate the mirage effect when it's hot and roads are looked at a grazing angle.
- Simulate optical defects of real cameras.
- Simulate humans
3D graphics for AI
"Dielectric Reliability in GaN MIS-HEMTs"
Faculty Advisor: Jesus del Alamo
Mentor(s): Ethan Lee
Contact e-mail: ethanlee@mit.edu
Research Area(s): Applied Physics, Materials and Devices
GaN based Metal Insulator Semiconductor High Electron Mobility Transistors (MIS-HEMTs) show great promise as the next generation technology for power control ranging from consumer electronics to satellites. The interface between GaN and AlGaN introduces a thin sheet of electrons called the 2-Dimensional Electron Gas (2DEG) that result in greatly enhanced current driving performance compared to Si based transistors. On the flipside, the presence of this 2DEG and the more complex heterostructures complicate understanding what is truly going on inside GaN MIS-HEMTs.
Your role will be to help design and carry out creative and insightful measurements to help understand the complex physics of GaN MIS-HEMTs with the ultimate goal of improving dielectric reliability issues that inhibit widespread commercial deployment.
Dielectric Reliability in GaN MIS-HEMTs
"Large Scale Machine Learning and Data Analytics on Graphs"
Faculty Advisor: Saman Amarasinghe
Mentor(s): Yunming Zhang
Contact e-mail: yunming@mit.edu
Research Area(s): Artificial Intelligence, Computer Systems
Recent years have seen the rise of many large graphs. Large scale social networks for recommendations used by Facebook and Twitter, knowledge graphs for information retrieval used by Google, transaction graphs for fraud detections used by Venmo and AntFinance, road graph for routing used by Uber and Lyft. Extracting useful information from these graphs and networks is a very important task.

Scalability is a big challenge to applying machine learning and data analytics techniques on these large graphs. Many of the recent advances in machine leaning on graphs, such as various graph centralities, collaborative filtering and graph neural networks are only tested on smaller graphs with tens of thousands of nodes. We want to scale these techniques (and develop new techniques if necessary) to graphs with billions of nodes and hundreds of billions of edges.

We are interested in finding students to develop new applications such as personalized recommendation in social networks, efficient queries on knowledge graphs and graph neural networks that can encode each node in the graph. The development will be done using GraphIt, a new high performance graph domain specific language, that holds the fastest published performance on many important graph applications, such as PageRank, BFS, Single Source Shortest Paths and Collaborative Filtering on large shared-memory machines. It is optimized for NUMA, cache.
Large Scale Machine Learning and Data Analytics on Graphs
"Efficient Deep Learning Hardware, Systems and Applications"
Faculty Advisor: Song Han
Mentor(s): Song Han
Contact e-mail: songhan@mit.edu
Research Area(s): Artificial Intelligence, Computer Systems
efficient inference: deep learning on the mobile and embedded devices for vision and speech. Link Link Link Link

efficient training: build large-scale distributed training systems that scales to hundreds of GPUs: Link

deep learning applications: learning to learn, and "do-it-yourself artificial intelligence" with these gadgets: Google Vision Kit: Link , and Google Voice Kit: Link

More on my website: Link
Efficient Deep Learning Hardware, Systems and Applications
"Pattern Mining for Network Traffic Management"
Faculty Advisor: Lizhong Zheng
Mentor(s):
Contact e-mail: lizhong@mit.edu
Research Area(s): Artificial Intelligence, Communications, Numerical Methods
We have recently developed a new approach, together with software packages, called "Multiple Alternating Conditional Expectation" (MACE). This is an algorithm that can be used to process high dimensional multi-modal data and extract low dimensional features that carries the most "important" information. We have shown theoretically that the approach has close connections to deep neural networks (NN), and can sometimes be used to replace neural networks or a part of a neural network.

This project is designed to test some of such new abilities. The specific task is to monitor network traffic, with data gathered from real-life cellular networks, and without looking into the data packets, as most of them nowadays are encrypted, we would hope to recognize certain patterns which can be used to guess the sender, the receiver, the contents, etc. about some of the network flows. This is an important functionality that the network equipment builders try to develop, and is a natural task for multi-modal data processing.

Applicant should be fluent in Python, and willing to understand some mathematical concepts.
"Investigating Privacy Preferences, Expectations and behaviors of voice activated devices"
Faculty Advisor: Ilaria Liccardi
Mentor(s):
Contact e-mail: ilaria@csail.mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
This project will focus on voice-activated devices, both physical (Google Home and Alexa) and mobile-based (Google Now and Siri). A student will be charged with looking into these devices and evaluating their technical ability to capture and retain data, by using traffic analysis and/or inspecting the actual hardware, as well as looking into the existing privacy practices of the companies that deploy these devices. The student will compare and evaluate them.

Most importantly, he or she will grapple with the more fundamental human behavioral questions related to:

- What kind of privacy perceptions and expectations have people considered and what are they concerned about when using these devices?
- What sorts of privacy and transparency controls are desirable for these types of device? In particular, how would people react if these indicators were added?
- How can we ensure that consumers fully understand the privacy trade-offs that come with the use of these technologies?

The right student will have a strong interest in privacy and human-computer interaction research; he or she will possess the technical knowledge or background needed to perform traffic analysis, to develop and deploy web-based probes (preferably Chrome extensions written in JavaScript) and mobile (Android or iOS) apps.
"3D electronics based on III-V semiconductors"
Faculty Advisor: Jesus del Alamo
Mentor(s): Alon Vardi
Contact e-mail: alonva@mit.edu
Research Area(s): Applied Physics, Materials and Devices
3D transistors based on III-V compound semiconductor are considered promising candidates for future logic applications. Due to the low effective mass in these materials, quantum size effects start already at 10 nm. With the advance Technology in our group this dimension range in within reach and we have fully developed FinFET and Nanowire processes reaching sub 10 nm dimension. To explore the transport properties in these extraordinary nano structures, we have developed nano transmission line model (TLM) structures and revolutionary nano Hall devices in which nano-electromagnet is embedded into the device. This device, by its own right may lead to breakthroughs in fast electronics. The scope of the project is wide and may include simulations, electrical measurements and process development – if you want to push the edge of electronic devices and go where no one has gone before - sign up today – extreme transistor group. 3D electronics based on III-V semiconductors
"Superscalar RISC-V Processors"
Faculty Advisor: Arvind
Mentor(s): Andy Wright
Contact e-mail: arvind@mit.edu
Research Area(s): Computer Systems
Superscalar processors differ from each other in their microarchitectures in terms of the resources they share (e.g., memory systems, store buffers, floating point units) and the policy they use in scheduling instructions. In this project, the student would explore the power-performance-area (PPA) of these choices by building several such processors starting from an in-order RISC-V processor. The designs must run on FPGAs and capable of booting Linux, as well as running a slew of benchmark suites. The designs should also be synthesized into ASICs to quantify area and timing properties.
If time permits we will also do a similar study with out-of-order processor whose non-superscalar code would be provided.
This project should lead to both to a publishable paper as well as generate material that can be used in architecture classes like 6.004, 6.175 and 6.823.
"Learning to Generate 3D Point Clouds"
Faculty Advisor: Justin Solomon
Mentor(s): Yue Wang
Contact e-mail: valianter.wang@gmail.com
Research Area(s): Artificial Intelligence, Graphics and Human-Computer Interfaces, Signals and Systems
Generation of 3D data has been attracting considerable interest for graphics/vision applications. Point clouds are a particularly popular shape representation among those looking to apply machine learning to this problem.

In this project, we will attempt to learn how to generate high-quality point clouds from image input. The first step will be to generate point clouds for small shapes, as studied in previous work (see the accompanying image, produced using PointNet). These generated point clouds can be used for virtual- or mixed-reality modeling. The second step will be to learn how to generate a realistic laser scan for an outdoor scene given a map of the scene as input. We try to answer the question: Can we simulate or generate a real point cloud that can’t be distinguished from real point clouds captured by LiDAR? If so, we can generate tons of training data for rare edge cases, needed to design perception system of self-driving cars. We are also happy to take guidance from our SuperUROP regarding applications and directions for this work.

Useful skills include:
* Understanding of machine learning and deep learning, preferably knowing and having experience with Tensorflow or PyTorch
* Basic knowledge of computer vision and computer graphics
* Familiarity with C++ or Python
* Enthusiasm to participate in this research project
Learning to Generate 3D Point Clouds
"Using Bitcoin to Support Online Journalism"
Faculty Advisor: David Karger
Mentor(s):
Contact e-mail: karger@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Tipsy ( Link ) is a project, funded by the Knight foundation, to explore a new way to support online Journalism using microdonations. Tipsy lives in your browser and periodically lets you make payments to news sites based on the amount of time you've spent reading them. Right now, Tipsy uses traditional credit card payments. I'd like to extend Tipsy to support payment by bitcoin, which would streamline the payment process.
There's also a possibility of extending this approach to support donation support for arbitrary creative commons licensed content.

Tipsy is a javascript browser extension for chrome; the idea candidate is familiar with Javascript web programming, and extension authoring.
"Squadbox"
Faculty Advisor: David Karger
Mentor(s): Amy Zhang
Contact e-mail: axz@csail.mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
We are building a tool to help people facing online harassment get help from friends. The tool helps people set up filters to forward stranger or potentially harassing messages to Squadbox, which then distributes the message to friends, who can perform actions such as approve it, file it away, tag it, redact portions, or summarize the email. The tool currently works with email though you could work on building it out to platforms like social media.

We are looking for UROPs with the following skills to continue developing Squadbox:
- (required) Front end web development (HTML, CSS, Javascript)
- (required) Back end web development (Python, Django, MySQL, email servers)

Possible additional skills or interests depending on what aspect of the tool you would like to work on:
- (optional) Experience connecting with APIs such as ones by Gmail, Twitter, Reddit, Youtube, etc.
- (optional) Experience or interest in security audits of our web stack, or building features related to maintaining user privacy
- (optional) Experience or interest in the integration of harassment machine learning detection models into Squadbox
- (optional) Interest in designing interventions to help moderators with burnout, or recipients of harassment with emotional stress

You can read more about the project here:
Our blog: Link
A Mozilla blogpost about us: Link
Our latest research paper: Link
Try out the tool itself: Link are building a tool to help people facing online harassment get help from friends. The tool helps people set up filters to forward stranger or potentially harassing messages to Squadbox, which then distributes the message to friends, who can perform actions such as approve it, file it away, tag it, redact portions, or summarize the email. The tool currently works with email though you could work on building it out to platforms like social media.

We are looking for UROPs with the following skills to continue developing Squadbox:
- (required) Front end web development (HTML, CSS, Javascript)
- (required) Back end web development (Python, Django, MySQL, email servers)

Possible additional skills or interests depending on what aspect of the tool you would like to work on:
- (optional) Experience connecting with APIs such as ones by Gmail, Twitter, Reddit, Youtube, etc.
- (optional) Experience or interest in security audits of our web stack, or building features related to maintaining user privacy
- (optional) Experience or interest in the integration of harassment machine learning detection models into Squadbox
- (optional) Interest in designing interventions to help moderators with burnout, or recipients of harassment with emotional stress

You can read more about the project here:
Our blog: Link
A Mozilla blogpost about us: Link
Our latest research paper: Link
Try out the tool itself: Link
"Assessing respiratory health via model-based processing of the capnogram"
Faculty Advisor: George Verghese
Mentor(s):
Contact e-mail: verghese@mit.edu
Research Area(s): BioEECS, Signals and Systems
Capnography is the measurement of CO2 concentration in exhaled breath, and is a widely used monitoring modality in various clinical settings. Previous work in our group has shown how the diagnostic power of capnography can be greatly extended by a more systematic and quantitative analysis than is typically performed. We have had some success with both machine learning as well as mechanistic modeling approaches (the figure depicts the latter). This project will focus on extending and refining the model-based approach, in order to enable application to: (a) distinguishing patients with congestive heart failure (CHF) from those with chronic obstructive pulmonary disease (COPD); and (b) monitoring severity and response to treatment in patients with asthma. We have a good base of clinical data already, with much more to be collected in the coming year from children with asthma at Boston Children's Hospital and from adults with CHF or COPD at Beth Israel Deaconess Medical Center. Ideal background would be 6.011 + 6.022 + a strong interest in clinical application of signals and systems approaches. Assessing respiratory health via model-based processing of the capnogram
"Predicting gene function via hyperbolic space embedding"
Faculty Advisor: Bonnie Berger
Mentor(s): Hoon Cho
Contact e-mail: bab@csail.mit.edu
Research Area(s): Artificial Intelligence, BioEECS
Gene function prediction is an important problem in computational biology, where our current knowledge of what different genes do in human is used to infer the properties of poorly-characterized genes. Better elucidating the functional roles of genes can lead to novel biomedical discoveries, such as finding new therapeutic targets for a disease.

Most existing approaches for this problem consider each functional category (e.g., "response to viral infection") as a separate prediction task. However, many categories currently contain only a small number of genes, which presents a significant hurdle for obtaining accurate prediction.

The aim of this project is to overcome this challenge by allowing information transfer between semantically-related categories to enhance the prediction of less-understood categories. We will newly explore the use of hyperbolic space as a medium to carry out such information transfer, in which the hierarchical structure of functional categories is more faithfully encoded by the distance metric compared to the commonly used Euclidean space.

This project will combine ideas from machine learning, statistics, math (geometry), and biology. Prior exposure to these subjects is preferred but not required. Familiarity with programming is required.
"From Mockup to Web App: Building the Next-Generation Web Template Language"
Faculty Advisor: David Karger
Mentor(s): Lea Verou
Contact e-mail: karger@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Think web frameworks like Node and Backbone are cool? Then help us develop the future of web templates. We are working on a web template language that continues to have benefits long after the page is rendered, including: rich copy-and-paste of data between websites, in-browser WYSIWYG editing, automatically-generated APIs, and site themes that are trivially transportable from site to site. We aim to empower casual web users with the ability to make professional web sites: from just a mockup, we hope to infer the data-backend and editing interface; by just pointing at another site, we hope to import that site's style for reuse on one's own. Experience with Javascript (or Coffeescript) and web development is a plus, as is good performance in 6.813/831. From Mockup to Web App: Building the Next-Generation Web Template Language
"Information Scraps, Quick Notetaking, and Personal Information Organization"
Faculty Advisor: David Karger
Mentor(s):
Contact e-mail: karger@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Our lives are filled with small, random scraps of information that seem to have no natural home. Where do we put them, and how do we find them later? We've created List.it (Link ), a fast, lightweight browser extension for capturing and organizing such scraps. Listit has had over 25,000 active users who have recorded more than 100,000 scraps. Analyzing them we've discovered important subpopulations such as packrats, minimalists, and spring cleaners. To advance our study of personal information organization, we want to study the activity of our current users (both the content they create and their interaction with it), add useful functionality to the tool (such as sharing, reminding, and context sensitive retrieval) and study the way users react to the new functionality. Information Scraps, Quick Notetaking, and Personal Information Organization
"Transparent Web Browsing"
Faculty Advisor: David Karger
Mentor(s):
Contact e-mail: karger@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Nowadays, all sorts of shady companies are collecting information about your browsing activities and using it for their own mysterious purposes. How could that information be used to your benefit? We propose to build Eyebrowse, a web browser extension that gathers information about your web browsing activities and shares that information (under your control) with others to mutual benefit. Potential applications include discovering interesting new web sites based on the browsing activity of your friends, improving web navigation by blazing trails to the important parts of web sites, supporting chance encounters when you and your friends are visiting the same web site, collaboratively browsing the web, identifying links between pages that ought to exist but don't, reporting on global web activity trends, tagging sites and pages according to the interests of people who visit them, and other exciting applications that you will come up with. Experience with Javascript, Django and general web application design is a plus, as is good performance in 6.813/831.
"The Future Textbook"
Faculty Advisor: David Karger
Mentor(s):
Contact e-mail: karger@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Now that we can put textbooks on the web, how can we change them to make them better? How can we make them more dynamic, more adaptable to individual students, more sociable, or more informative? We've tackled some of these questions with Nb ( Link
), a tool that lets students hold forum-type discussions in the margins of their online reading material. Nb is currently in use in roughly 25 classes at 6 universities. We have a long list of improvements to implement and assess in Nb, including social moderation, key-question highlighting, organization via tagging, chat and wiki functionality, support for sketching diagrams and other non-text annotations, hot-spot mapping for faculty, html and video annotation, and many others that students like you think of. Experience with Javascript and Django is a plus, as is good performance in 6.813/831.
The Future Textbook
"Fabrication and Characterization of Thin Film Superconductors and Nanowires"
Faculty Advisor: Karl K. Berggren
Mentor(s): Andrew E. Dane
Contact e-mail: aedane@mit.edu
Research Area(s): Applied Physics, Materials and Devices, Nanotechnology
The goal of this project is to deposit and characterize few-nanometer-thick superconducting films which are the raw material used in the fabrication of sensitive photonic devices, especially superconducting nanowire single photon detectors (SNSPDs). The performance of SNSPDs is dependent on the material used to make them. At the moment, two main materials are used to fabricate SNSPDs: niobium nitride (NbN) and tungsten silicide (WSi). Devices based on NbN are typically faster, while devices based on WSi usally have better detection efficiency, among other differences. Recently, Professor Berggren’s group has published work on making SNSPDs from bilayers of WSi and NbN.

In this project, a student will investigate the superconducting properties of few-nm-thick bilayer films and wires fabricated on them. By composing stacks of different composition we are able to (1) vary the superconducting critical temperature of the film stack by varying the relative amount of each material, (2) investigate the effect of differing surface layer on the Tc and overall film roughness, (3) improve the robustness of a given material to fabrication by coating it in a protective metal or dielectric layer (such as self-passivating Ti).
"Leveraging Clinical Data Sets to Optimize Oxygen Delivery to Newborns"
Faculty Advisor: Thomas Heldt
Mentor(s): Dr. Wendy Timpson
Contact e-mail: thomas@mit.edu
Research Area(s): Artificial Intelligence, BioEECS
Tight oxygen titration in the preterm neonate is a key aspect of neonatal intensive care due to the mortality associated with hypoxia (low oxygen saturation) and the morbidity associated with hyperoxia (high oxygen saturation) in this vulnerable population. Despite these known complications of sustained oxygenation outside target ranges, most Neonatal Intensive Care Units (NICUs) fail to reliably maintain infants’ oxygenation saturations within target range. The primary goal of this project is to leverage large volumes of physiological data streams collected in the NICU to identify clinical, demographic, physiological and workflow factors that place preterm infants at risk for hypoxia and hyperoxia.

This project offers an opportunity to partner with neonatologist and clinical staff at Beth Israel Deaconess Medical Center and actively participate in data analytics to directly improve the care of the tiniest patients.
"Unobtrusive estimation of blood pressure variation"
Faculty Advisor: Thomas Heldt
Mentor(s): Daniel Teichmann
Contact e-mail: teichi@mit.edu
Research Area(s): BioEECS, Circuits, Signals and Systems
Time difference techniques for the estimation of blood pressure (BP) are based on the time delay between heart contraction and pulse arrival in the periphery (pulse arrival time (PAT), measured with one ECG and one PPG sensor) or the time a pulse waves needs from one point on the arterial branch to another (pulse transit time (PTT), measured with two PPG sensors). Both methods have shown to be in correlation with BP variation.
The aim of this project is to derive estimations of the BP variation based on PAT by using a novel unobtrusive sensing technique instead of a conventional ECG sensor. This unobtrusive sensor technique is called magnetic induction monitoring and allows measurement of vital signs in a noncontact way even through several layers of clothing.
The course of this project is separated into a) the building of a magnetic induction sensor (based on already existing electrical layouts), b) conducting measurements including test protocol definition, setup of the experiment and actual data recording, and c) analysis of the data.
The latter will necessitate the use and adaption of physiological models describing the link between pulse wave velocity and BP via (among others) arterial stiffness, diameter and vasoconstriction.
Unobtrusive estimation of blood pressure variation
"A sensor fusion algorithm for reliable estimation of physiological parameters"
Faculty Advisor: Thomas Heldt
Mentor(s): Daniel Teichmann
Contact e-mail: teichi@mit.edu
Research Area(s): BioEECS, Signals and Systems
This project aims at the development and validation of an algorithm for the extraction of heart rate and respiratory rate from noncontact sensor signals.
In a clinical study a novel noncontact measurement device was tested on 20 patients during dental treatment. During the entire measurement a gold standard electrocardiogram, photoplethysmogram and a respiration effort signal were recorded as pulse and respiration reference, respectively.
The applied noncontact measurement device comprises two different sensor modalities each placed at three different locations within the backrest of the dental treatment unit (6 channels). Both sensor modalities are able to derive cardiac and respiratory activity in a noncontact way. The signal quality is different among the channels. This difference exists not only inter- but also intra-individually and may also change over time. Next to the occurence of motion artifacts due to the dental treatment, another difficulty is posed by the fact that the transmission functions between the signals' sources (heart and lung motion) and the recorded signals vary over time, i.e. signal morphology and phase might vary. The final algorithm should implement a method for signal fusion which is able to maximize the coverage rate of estimated parameters while maintaining a low estimation error.
A sensor fusion algorithm for reliable estimation of physiological parameters
"Learning Brownian Motion Trees and Applications to Cell Differentiation"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): BioEECS, Numerical Methods, Signals and Systems, Theoretical Computer Science
Every cell in our body contains the same genetic information. However, we have many different cell types and they all show different gene expression patterns. In order to understand how this is possible, we will study how the different gene expression patterns develop, starting from a stem cell and differentiating into the different cell types. We will assume that gene expression develops along the tree according to a multivariate Brownian motion with correlation among the genes. Given gene expression data from various points along the unknown differentiation tree, the goal of this project is to develop algorithms that can simultaneously map the cells to the differentiation tree and learn the underlying tree topology. What statistical guarantees can be obtained? How well does the algorithm perform on simulations? Does it provide meaningful differentiation trees when applied to real biological data?
"Ellipsoid Packing and Applications to Chromosome Organization"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): Applied Physics, Artificial Intelligence, BioEECS, Control, Numerical Methods, Signals and Systems, Theoretical Computer Science
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 use optimization theory to develop algorithms that find locally optimal minimal overlap configurations of ellipsoids under various constraints. Biologically relevant constraints include distance to the boundary of the container, overlap between specific ellipsoids, or changing container shapes. The obtained ellipsoid configurations will be compared to experimental data and used to predict the new chromosome neighborhoods when altering the shape of the nucleus.
"Learning Causal Graphs and Applications to Gene Regulation"
Faculty Advisor: Caroline Uhler
Mentor(s):
Contact e-mail: cuhler@mit.edu
Research Area(s): Artificial Intelligence, BioEECS, Control, Numerical Methods, Signals and Systems, Theoretical Computer Science
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?
"A memory hierarchy for modern programming languages"
Faculty Advisor: Daniel Sanchez
Mentor(s): Po-An Tsai
Contact e-mail: poantsai@csail.mit.edu, sanchez@csail.mit.edu
Research Area(s): Computer Systems
Current computers use abstractions designed for ancient and unsafe programming languages, like C and Fortran. In particular, memory hierarchies expose a flat address space and rely on expensive virtual memory mechanisms to provide protection and isolation among programs. But modern languages, like Go, Java, and Javascript, rely on automatic memory management and disallow raw memory accesses to eliminate many types of memory-access bugs. While productive, these languages are mismatched with current abstractions: they need expensive runtime support to manage their flat memory address space (e.g., garbage collection), and virtual memory is overkill for them and incurs unnecessary overhead.

We have recently designed a new type of memory hierarchy and hardware/software interface tailored to support modern languages with automatic memory management. This system removes most of the overheads of conventional memory systems: the system moves data across a hierarchy of memories, but does so without caches of virtual memory. This systems also removes the overheads of memory allocation and garbage collection, and maintains memory safety.

This new memory hierarchy has opened up multiple research opportunities, each of which can be a SuperUROP project:

- Because it has no caches, this new system is immune to the recently-announced Spectre and Meltdown attacks. You could analyze the security properties of this new system and prove that it is immune to other timing side-channel attacks.

- You could design new resource management mechanisms to divide the memory hierarchy across the programs sharing the system to improve performance.

- Our system makes it possible to transform objects on the fly, as they move through the memory hierarchy. You could implement support for compressed objects and demonstrate their benefits vs. conventional main-memory compression techniques.
"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 Systems
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, 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, SCC, which automatically parallelizes sequential C/C++ applications.

- Help develop and scale Swarm's FPGA implementation, 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
"Efficient Cryptography for the Internet of Things"
Faculty Advisor: Anantha Chandrakasan
Mentor(s): Utsav Banerjee
Contact e-mail: anantha@mit.edu
Research Area(s): Circuits, Computer Systems
The IoT has introduced a vision of the Internet where all computing and sensing devices are interconnected. However, these devices also have the potential to become attractive targets for cyber attackers. Therefore, it is important to ensure that these devices remain secure while adhering to strict power budgets.

Our group has designed a chip that couples a low-power RISC-V micro-processor with energy-efficient cryptographic accelerators for AES, SHA and ECC. This platform can be used to evaluate multiple aspects of IoT security:
1. Side-channel analysis of standard cryptographic algorithms
2. Implementations of light-weight cryptography
3. Analysis of next-generation elliptic curves
4. Software benchmarking of post-quantum public key algorithms
5. System-level energy modelling of security protocols

Prerequisites: Experience with programming in C
Efficient Cryptography for the Internet of Things
"Trans-Continental Telemedicine"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Artificial Intelligence, Communications, Computer Systems, Graphics and Human-Computer Interfaces, Signals and Systems
Emory Healthcare has recently transferred some of its doctors and nurses to Australia so that they can work there during day time to provide better services to patients during night time in Atlanta. Patients can benefit from using a three-pronged approach that incorporates on-site personnel, off-site personnel, and data analytics. This project involves study of these building blocks leading to a concept demonstration prototype. While other students are welcome to apply, preference will be given to students currently taking 6.884 for credit.
"Clinical Decision Support System"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Artificial Intelligence, Computer Systems, Graphics and Human-Computer Interfaces
Non-experts increasingly prescribe expensive and risky interventions in medicine. This is particularly true in complex, acute care settings where numerous allied health personnel interact to make rapid decisions. While the first generation of Clinical Decision Support (CDS) Systems marginally reduces the use of low value interventions, such guidance is frequently over-ridden as it is perceived as being an impediment to patient care by providers at the bedside, especially during critical situations. Based on the three-pronged approach to healthcare articulated by the project supervisor, this project involves research, development, and implementation of the prototype of a new version of CDS that will incorporate a human element by engaging domain experts using real-time mobile applications and dashboard technologies. This implementation will provide real-time expert opinion to providers and demonstrate improved compliance, reduced risk to patients, and overall, better care at lower cost.
"Interoperability of Healthcare Aids, Devices, Communication and Information Systems"
Faculty Advisor: Amar Gupta
Mentor(s):
Contact e-mail: agupta@mit.edu
Research Area(s): Circuits, Computer Systems, Control, Materials and Devices, Signals and Systems
The broad application of new digital health technologies is contingent on better ways to share data between EHR, IOT, pharmacy, and lab systems. This project will focus on developing a practical approach that can be deployed around the world. Preference will be given to students who are taking 6.884 this term; others are welcome to apply too.
"(Re)Thinking Software Design"
Faculty Advisor: Daniel Jackson
Mentor(s):
Contact e-mail: dnj@mit.edu
Research Area(s): Computer Systems
Our goal is to develop a theory of software design focused on the conceptual structure that underlies software. Key threads include developing suitable diagrammatic notations; articulation of design criteria; and reworking existing applications like Git.
--
For decades, 'software design' has been about the internal structure of the code. Recently though, with the growing interest in design thinking, practitioners and researchers have started focusing more on the design of the behavior of the software. The behavior not only determines the user experience, but also shapes the implementation in fundamental ways. In this project, we are developing a radical new approach to software design that focuses on this aspect. The key idea is the identification of conceptual constructs that embody the key behavioral features of the application. To realize the goal of a new design method for software, we analyze existing systems and applications and develop new ones, polishing our design theory in response to what we learn from case studies.

See: Link
"Déjà Vu: Constructing Apps from Clichés"
Faculty Advisor: Daniel Jackson
Mentor(s): Santiago Perez De Rosso
Contact e-mail: sperezde@csail.mit.edu
Research Area(s): Computer Systems
Déjà Vu is a new platform for end-user development of apps with rich functionality. It features a novel theory of modularity for binding concepts; an extensive library of reusable concepts; and a WYSIWYG tool for specifying bindings and customizing visual layout

--
As a user you might have noticed the fundamental similarities between the many applications you use daily. Maybe it was the day you were scrolling through your Facebook news feed and then through your Twitter feed? Or when you gave a 5-star review to a restaurant in Yelp, and then to a book in Amazon? Or that time when you replied to a tweet and found yourself later replying to a comment on Reddit? Now picture the many software engineers developing web applications, ranging from internal business applications to those used by millions of users. How many of these engineers are, at this moment, working on implementing a password recovery mechanism? How many are adding some kind of news feed to their application? What about adding chat functionality? A shopping cart? Adding star ratings or likes? Letting users write comments? Surely enough, in each of these instances, developers are not all doing the exact same thing. In some cases, the feed is listing posts authored by users, in other cases it’s showing shopping products, or books. Some developers need the feature to be tweaked in a unique way, or are using different languages and frameworks. But it is the premise of this project that, fundamentally, they are all doing the same thing: combining pre-existing concepts in novel ways. And that if we could successfully exploit this fact, applications could be built much faster than how they are built today.

See: Link
"An Interlock for Self Driving Cars"
Faculty Advisor: Daniel Jackson
Mentor(s):
Contact e-mail: dnj@mit.edu
Research Area(s): Computer Systems
Self-driving cars are likely to be safer, on average, than human-driven cars. But they may fail in new and catastrophic ways that a human driver could prevent. This project is designing a new architecture for a highly dependable self-driving car.

Human drivers are good at avoiding huge boulders that come into view a mile away; at adjusting their speed when they can no longer see well; and can often even stop suddenly or pull over when continuing would be dangerous. Self driving cars generally can't do any of these things. In this project, we are developing a new architecture to bring this kind of flexibility to self-driving cars.

The keys ideas are:

* A reconfigurable network that allows sensors to be reassigned if some fail, a control algorithm to be replaced by a simpler one, and computational resources to be increased for select software components;

* A system modeler that maintains an explicit representation of the state of the system -- software and hardware -- and detects when it goes outside the envelope in which it was designed to operate;

* An assurance case that is constructed statically from the design to determine the operating envelopes of different controllers, and that can be monitored on the fly.

See: Link
"Building a Principled Science of Deep Learning"
Faculty Advisor: Aleksander Madry
Mentor(s):
Contact e-mail: madry@mit.edu
Research Area(s): Artificial Intelligence, Theoretical Computer Science
The project aims to develop a principled understanding of deep learning from an optimization and robustness perspective. Specifically, its main focus will be on an experimental study that aims to identify some of the underlying principles of deep learning. These identified principles will then guide the development of theoretical models and, in turn, lead to improved understanding of this intriguing domain as well as new tools and techniques.

Two topics of focus will be: improving the robustness of deep learning models; and understanding the optimization landscape of deep neural network training.

Fair warning: This project is challenging. It will require good implementation skills as well as strong algorithmic background and mathematical maturity. Basic knowledge of continuous optimization is a plus.
Building a Principled Science of Deep Learning
"Automated electrochemical sensors for point-of-care monitoring"
Faculty Advisor: Joel Voldman
Mentor(s):
Contact e-mail: voldman@mit.edu
Research Area(s): BioEECS, Circuits
This project focuses on using embedded systems to create automated and portable medical sensors. Handheld sensing of biological molecules can transform medicine. Think of blood glucose monitoring: handheld meters measure blood glucose from less than a drop of blood in just a few seconds. We are developing sensors to measure other molecules, and are interested in automating the production and testing of the sensors. Since the sensors are electrical in nature, and involve microfluidic flows, we want to use an embedded controller to control the electronics and fluidics, in turn enabling fully automated and portable control. Automated electrochemical sensors for point-of-care monitoring
"Computational analysis of immune status"
Faculty Advisor: Joel Voldman
Mentor(s):
Contact e-mail: voldman@mit.edu
Research Area(s): Artificial Intelligence, BioEECS
This project focuses on using learning-based methods to infer the function of white blood cells. White blood cells are central players in immunity. As such, counting and classifying white blood cells is one of the most common medical tests. For example, infections are associated with spiking of white blood cell counts. Beyond counting, there is a desire from clinicians to understand the function of these cells, and as a result many tests have been developed to test how these cells behave. Unfortunately, these tests are fairly laborious and costly. Can we obtain similar information from images? This project will attempt to find out. Computational analysis of immune status
"Sensors for Oral Health"
Faculty Advisor: Joel Voldman
Mentor(s):
Contact e-mail: voldman@mit.edu
Research Area(s): BioEECS, Materials and Devices
This project will focus on making wireless sensor systems that can be used to measure oral health. When you go to the dentist, many of the tools they use have not changed in decades. For example, assessment of gum disease is done by visual inspection (do gums bleed during cleaning) or depth measurements via a metal pick. Instead, we want to create electrical sensors that can measure gum disease at a molecular level. This will involve sensor design, some wet lab chemistry, electronics reader design, and maybe even wireless device prototype. Sensors for Oral Health
"Polar Factorization of Maps via Optimal Transport"
Faculty Advisor: Justin Solomon
Mentor(s): Ed Chien
Contact e-mail: edchien@mit.edu
Research Area(s): Numerical Methods, Theoretical Computer Science
A square matrix may be decomposed as a product A = SR, of a symmetric positive definite matrix S and an orthogonal (rotation) matrix R. Amazingly, Brenier observed that this fact may be generalized to nonlinear maps from d-dimensional space to itself, with optimal transport. One may uniquely write such a map, f(x) = s(r(x)), as a composition of a measure-preserving map r and an optimal transport map s. The map r is characterized as the closest measure-preserving map to f (in some rigorous sense).

In this project, we aim to numerically calculate this map and explore applications of this decomposition in manipulation of smoke simulations and modelling of transport problems with non-uniform costs. The student will gain an introduction to the hot field of numerical optimal transport and first-hand experience with application of basic numerical analysis and optimization techniques.
Polar Factorization of Maps via Optimal Transport
"Understanding the Topology of SO(3)/O Through Computation"
Faculty Advisor: Justin Solomon
Mentor(s): Paul Zhang
Contact e-mail: pzpzpzp1@gmail.com
Research Area(s): Graphics and Human-Computer Interfaces, Numerical Methods, Theoretical Computer Science
The goal of our research is to understand the topology of frame fields used in hexahedral meshing. Hexahedral meshing is the task of decomposing a volume into many smaller cuboid components, often used for physical simulation. Recent methods employ the use of frame fields which assign three smoothly varying orthogonal vectors to every point in the volume. Frames are often represented by general 3D rotations (rotation matrices, quaternions, axis-angle, etc), however these representations do not capture the cubical symmetries of a frame. The goal of this project will be to understand the reduced space SO(3)/O, and to compute an exact representation of it.

Approach: SO(3)/O can be described as the set of rotations that are closer to the origin than to any other octahedral rotation. This can be calculated by creating a Voronoi diagram with centroids placed at octahedral rotations. A complication arises however since bisecting planes between two rotations are not flat. The task will be to compute the curved bisecting plane and mesh the Voronoi cell that results.

Prerequisites: The student must be proficient with Matlab or an alternative programming language. This position is well suited for candidates with CS/Math background. Familiarity with computation on meshes is preferred but not necessary.
Understanding the Topology of SO(3)/O Through Computation
"Sampling in the Space of Redistricting Plans"
Faculty Advisor: Justin Solomon
Mentor(s): Sebastian Claici
Contact e-mail: sclaici@mit.edu
Research Area(s): Artificial Intelligence, Numerical Methods, Signals and Systems, Theoretical Computer Science
To combat gerrymandering, we must propose alternatives to current districts that have nice geometrical properties, are fair, and obey civil rights law. One starting point is to uniformly sample from the space of redistricting plans. We are looking for a UROP to research and develop novel methods for sampling redistricting plans that are based on state-of-the-art graph algorithms including random spanning tree generation and fast maximum flow implementations. The student will be guided and exposed to research in geometry, topology, and political redistricting.

Prerequisites: The student should have a good understanding of classical graph algorithms and the ability to implement them in a programming language of their choosing.
Sampling in the Space of Redistricting Plans
"Distortion-minimizing bijective surface correspondence"
Faculty Advisor: Justin Solomon
Mentor(s):
Contact e-mail: jsolomon@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces, Numerical Methods, Theoretical Computer Science
The problem of *correspondence*, or mapping from one surface into another, finds application in medical imaging, computer graphics, CAD, and other disciplines.

State-of-the-art bijective correspondence involves mapping two surfaces into a common domain like the plane and then composing these maps. This guarantees that every point gets mapped, but it does *not* minimize distortion (stretch/shear) of one surface mapped onto the other. In this project, you will develop tools for distortion-minimizing correspondence using these methods as a starting point.
Distortion-minimizing bijective surface correspondence
"Learning How to Project Your Voice with VR"
Faculty Advisor: Tony Eng
Mentor(s):
Contact e-mail: tleng@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
“Can you hear me?” You might ask this at the start of your presentation to see if you are loud enough to be heard. If a person isn’t loud enough, they might start off loud, but they will quickly return to their original (low) volume. Is there a way that we can help students work on volume (apart from saying “speak louder”) by helping the student recalibrate how loud they sound to themselves?

The proposal is to do so by building a VR game in which the student is speaking in a room and there is a visual indicator of how far their voice projects.

The main tasks are:

1. Create a game with various (unconnected) speaking venues using the Oculus Rift
2. Given a player’s volume level, figure out how far away the player can be heard
3. Determine how to visually display that to the player
4. Make it all into a game that can include other ideas for helping the player develop projection/volume
5. Test the game on various students who wish to work on volume and projection to determine if and when there is any corresponding improvement in volume.

Preference to students who have worked with the Oculus Rift before.
"A Depth Estimation Platform for Time-of-Flight Imaging"
Faculty Advisor: Vivienne Sze
Mentor(s): James Norkay
Contact e-mail: jnoraky@mit.edu
Research Area(s): Artificial Intelligence, Computer Systems
Description: Depth sensing is useful in a variety of applications that range from robotics to commercial robotics. Time-of-flight (TOF) cameras obtain dense depth measurements, known as a depth map, by emitting light and measuring its round trip time. However, the illumination source of a TOF camera is often power hungry and can limit the battery life for mobile applications. One way to address this is to lower the rate at which depth maps are acquired and use RGB images, which can be efficiently and concurrently collected, to estimate depth without illuminating the scene. The goal of this project is to prototype a system that estimates depth by incorporating RGB images and partial depth information. This project will involve:
** Writing software to interface with commercial TOF sensors
** Building a GUI to visualize the estimated depth maps
** Evaluating the impact of the estimated depth map in various algorithms that use depth
** Embedded software/FPGA development (e.g. 6.111)
"Linguistic Analysis of Wikipedia for Question Answering"
Faculty Advisor: Boris Katz
Mentor(s): Sue Felshin
Contact e-mail: boris@csail.mit.edu
Research Area(s): Artificial Intelligence
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, we hope to considerably increase the precision and coverage of our START Natural Language Question Answering System. The project will involve our WikipediaBase system which turns Wikipedia infoboxes and other semi-structured Wikipedia data into a virtual database accessible via natural language.
"Hydration Monitoring Device"
Faculty Advisor: Luca Daniel
Mentor(s): Ian Butterworth
Contact e-mail: luca@mit.edu
Research Area(s): Applied Physics, BioEECS, Circuits
Our hydration project (hydration.mit.edu) is focused on developing the technology missing for reliable and physiologically meaningful hydration tracking, with an aim of optimising hydration and avoiding dehydration. We created the project at MIT with a primary focus on the substantial need for improved hydration management in the elderly, but with a view to the needs in many sectors and the broader population.
We are currently carrying out clinical testing in Madrid and Boston of two novel non-invasive technologies in a prototype.
Hydration Monitoring Device
"Predictive Model of Urban e-Mobility integrated with Renewable Energy Sources"
Faculty Advisor: Luca Daniel
Mentor(s): Michela Longo
Contact e-mail: luca@mit.edu
Research Area(s): Control, Energy, Graphics and Human-Computer Interfaces, Numerical Methods
Latest technological developments and renewed attention to eco-sustainability have fostered the vision of smart power grid interacting with smart cities. This revolution is now quickly spreading also to the field of mobility (e.g. human transportation), originally sustained exclusively by fossil fuels. This project aims at developing a predictive model and a software tool that will enable the increase adoption of electrical vehicles while leveraging renewable energy sources to help maintain the power grid stable. The model will be developed and calibrated using available data (e.g. the electric infrastructure, the current number of vehicles, population density, wealth and attitude towards technological innovation). The tool will be able to predict the impact of electric vehicles usage and charging station locations on electric power distribution network. Hence users will be able to explore tradeoffs and “what if” scenarios for instance with the objective of optimal placement of private and public charging stations. More specifically, our model will exploit the fact that electric vehicles may work either as loads (requiring energy from the grid and possibly absorbing surplus from local renewables), or as generators (feeding energy back into the grid, thus supporting the supply network in case of micro-shortages, especially at the local level). Predictive Model of Urban e-Mobility integrated with Renewable Energy Sources
"Evaluating the Robustness of Neural Networks"
Faculty Advisor: Luca Daniel
Mentor(s): Tsui-Wei (Lily) Weng
Contact e-mail: luca@mit.edu
Research Area(s): Artificial Intelligence, Numerical Methods
Although neural networks are becoming the core engine for driving Artificial Intelligence and Machine Learning research and technology at an unprecedented speed, recent studies have highlighted their lack of model robustness to adversarial attacks, giving rise to new safety, security, and socio-economical implications. In order to address this emerging issues, this proposal aims to provide a certified robustness evaluation framework that jointly takes into consideration an arbitrary neural network model and its underlying datasets. Specifically, we aim at developing an attack-agnostic robustness metric to evaluate the robustness of neural network classifiers. We further aim at providing efficient data-driven schemes to improve model robustness by pinpointing exemplary anchor points inferred from the underlying datasets. [just as a simple illustrating example, by making imperceptible changes to the image, neural networks can be fooled into misclassifying the ostrich in figure as a vacuum cleaner, a safe or even a shoe shop...] Evaluating the Robustness of Neural Networks
"Training Algorithms for the COIN"
Faculty Advisor: Cardinal Warde
Mentor(s):
Contact e-mail: warde@mit.edu
Research Area(s): Artificial Intelligence, Circuits, Computer Systems, Control, Signals and Systems, Theoretical Computer Science
The Photonic Systems Group is developing a Compact Opto-electronic Integrated Neural (COIN) network co-processor. The COIN employs arrays of photodetectors and thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. One of the key hardware subsystems is a dense 2-D array of bistable optical devices. We are exploring the development of such a subsystem, first at the discrete component level, by driving an array of LEDs with an array of photodetectors and thresholding electronics. This project involves: (1) design and modeling of the components and the associated bistable optical subsystem,(2) assembly and testing of a hybrid bistable optical array and (3) development and implementation of training algorithms for the COIN. Follow-on work would focus on the design and fabrication of the final compact VLSI integrated-circuit version of the COIN. Training Algorithms for the COIN
"Assembly of Hybrid COIN"
Faculty Advisor: Cardinal Warde
Mentor(s):
Contact e-mail: warde@mit.edu
Research Area(s): Circuits, Computer Systems, Materials and Devices, Signals and Systems
The Photonic Systems Group is developing a Compact Opto-electronic Integrated Neural (COIN) network co-processor. The COIN employs arrays of photodetectors and thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. One of the key hardware subsystems is a dense 2-D array of bistable optical devices. We are exploring the development of such a subsystem, first at the discrete component level, by driving an array of LEDs with an array of photodetectors and thresholding electronics. This project involves: (1) design and modeling of the components and the associated bistable optical subsystem,(2) assembly and testing of a hybrid bistable optical array and (3) development and implementation of training algorithms for the COIN. Follow-on work would focus on the design and fabrication of the final compact VLSI integrated-circuit version of the COIN. Assembly of Hybrid COIN
"2-D Bistable Optical Array"
Faculty Advisor: Cardinal Warde
Mentor(s):
Contact e-mail: warde@mit.edu
Research Area(s): Materials and Devices
The Photonic Systems Group is developing a Compact Opto-electronic Integrated Neural (COIN) network co-processor. The COIN employs arrays of photodetectors and thresholding circuits in silicon, in combination with arrays of light emitters and holographic optical interconnection elements sandwiched together into a brick. One of the key hardware subsystems is a dense 2-D array of bistable optical devices. We are exploring the development of such a subsystem, first at the discrete component level, by driving an array of LEDs with an array of photodetectors and thresholding electronics. This project involves: (1) design and modeling of the components and the associated bistable optical subsystem,(2) assembly and testing of a hybrid bistable optical array and (3) development and implementation of training algorithms for the COIN. Follow-on work would focus on the design and fabrication of the final compact VLSI integrated-circuit version of the COIN. 2-D Bistable Optical Array
"3D Printing + Context Aware Computing and Integrated Sensors"
Faculty Advisor: Prof. Stefanie Mueller
Mentor(s):
Contact e-mail: stefanie.mueller@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
We will build a piece of software that will analyze in which context users use their products to enable the next generation of product design and customization. For instance, imagine the difference between a construction worker wearing a watch and an office worker wearing a watch. The watch of the construction worker has to survive many external influences: the sunshine might degrade the colors of the wristband, and the heavy duty work might damage the casing. In contrast, the office worker might prefer a lighter watch and one that doesn't press against the wrist while typing on the computer.

We will first collect usage data using sensors integrated into the devices, such as the GPS from the watch, to analyze how and where users are using their devices. We will then create a user profile from this data and integrate this into an online shopping platform. When users go online to buy their next product, we will look at their profile and provide them with recommendations which product to buy based on their usage data.

We will combine this platform with 3D printing and parametric design. For instance, a parametric design of a watch might change the thickness and the material of the wristband automatically when outside and heavy duty use of the watch is detected.

Ideally you have done some basic electronics, signal processing of sensor data, and parametric design. However, we will also be happy to get you started with skills we have in the lab.

Link

All of our projects are planned as paper publications for the ACM CHI or ACM UIST conferences.
3D Printing + Context Aware Computing and Integrated Sensors
"3D Printed Electronics + 3D Modeling + Mesh Processing to Enable Next Generation of Technical Innovators"
Faculty Advisor: Prof. Stefanie Mueller
Mentor(s):
Contact e-mail: stefanie.mueller@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
We are building a design software that allows users to quickly generate custom PCB designs that fit into existing 3D models.

Today, it is becoming increasingly easier for people to use 3D printing to create their own objects. However, so far most objects are non-interactive since it is difficult to include sensor circuits that match a user's 3D design.

Plug-and-play sensors typically come in rectangular form factors that often do not fit into 3D models (e.g. Link Custom PCBs in different form factors would solve the problem, however, they are difficult to create for beginner users.

In this project, we will create a piece of software that analysis the shape of a 3D model and then automatically creates a PCB design that fits inside that shape. We will then fabricate this PCB on a low-cost PCB milling machine (e.g. Link to demonstrate this simple end-to-end workflow for beginners.

As an extension, we will play around with the Voxel 8 3D printer for 3D printed electronics to investigate if we can print the entire design with circuits in one go.

Ideally you have a background in 3D modeling / mesh processing and know something about electronics--but we will also be happy to get you started with the skills we already have in our group.

Link

All of our projects are planned as paper publications for the ACM CHI or ACM UIST conferences.
3D Printed Electronics + 3D Modeling + Mesh Processing to Enable Next Generation of Technical Innovators
"3D Printing + Electronics + Parametric Modeling for Adaptive Learning for Physical Tools"
Faculty Advisor: Prof. Stefanie Mueller
Mentor(s):
Contact e-mail: stefanie.mueller@mit.edu
Research Area(s): Graphics and Human-Computer Interfaces
Adaptive learning systems are already widely popular in learning math and language skills: if the learner performs the task correct the learner gets a harder task, if the learner makes a mistake additional support material is being displayed and the next task is easier.

In this project, we apply the concept to learning motor skills, such as learning how to bike. Rather than simply having training wheels that can either be attached or taken off, we monitor the learner's progress using sensors on the bike and then gradually move the training wheels inwards using motors to steadily increase the difficulty level until they are no longer required. Underlying our system is a parametric design software (Rhino/Grasshopper).

We need your help to extend our parametric design software and to build more examples to illustrate the concept. You will be free to choose which component you would like to focus on, just talk to us.

Link

All of our projects are planned as paper publications for the ACM CHI or ACM UIST conferences.
3D Printing + Electronics + Parametric Modeling for Adaptive Learning for Physical Tools
"Understanding language to see the world, and seeing the world to understand language"
Faculty Advisor: Boris Katz
Mentor(s): Andrei Barbu
Contact e-mail: boris@csail.mit.edu
Research Area(s): Artificial Intelligence
Why is human vision so much better than machine vision? How do we describe what we see? How do you recognize an event that someone is describing? How do you learn language when all you hear are sentences with words you don't understand?

We are interested in understanding these questions by developing models that jointly address vision and language. These models should allow you to perform tasks that come naturally to children but are difficult for machines: describing what you see, seeing what is being described, asking questions about what you see, determining if what you are seeing is reasonable, using knowledge from language to change your mind about what you are seeing, etc.

This project aims to develop new models, extend current models to novel environments, or answer novel questions about language and vision.
"Molecular Clock: Ultra-Stable Frequency Reference On A CMOS Chip"
Faculty Advisor: Ruonan Han
Mentor(s):
Contact e-mail: ruonan@mit.edu
Research Area(s): Applied Physics, Circuits, Materials and Devices
Polar gas molecule has narrow and stable transition lines in the sub-THz range. By probing the exact frequencies of these lines using a CMOS silicon chip, we can build a very small “atomic clock” that provides highly stable output frequency (<1E-11 error). Students in this project will be in charge of (1) testing the stability of the clock prototype and its sensitivity to temperature change, and (2) build a micro-controller circuit that senses the environment temperature and adjusts the clock’s output frequency accordingly to reduce its temperature dependency (1E-11~1E-12/degree). The student also has the chance to participate in the on-chip electronic design of this clock.

Prerequisite: analog and digital circuits (6.002).
Molecular Clock: Ultra-Stable Frequency Reference On A CMOS Chip
"Autonomous Driving"
Faculty Advisor: Daniela Rus
Mentor(s):
Contact e-mail: rus@csail.mit.edu
Research Area(s): Artificial Intelligence, Control
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. Autonomous Driving
"Printable Robots"
Faculty Advisor: Daniela Rus
Mentor(s):
Contact e-mail: rus@csail.mit.edu
Research Area(s): Artificial Intelligence, Graphics and Human-Computer Interfaces, Materials and Devices
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. Printable Robots
"Soft Robots"
Faculty Advisor: Daniela Rus
Mentor(s):
Contact e-mail: rus@csail.mit.edu
Research Area(s): Artificial Intelligence, Materials and Devices, 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):
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.
"Faster DNNs in TensorFlow"
Faculty Advisor: Saman Amarasinghe
Mentor(s): Riyadh Baghdadi
Contact e-mail: baghdadi@mit.edu
Research Area(s): Artificial Intelligence, Computer Systems
The COMMIT group (Link is developing a new code optimization framework for DSL (Domain Specific Language) compilers called Tiramisu. It takes high level code generated by a DSL compiler, optimizes it and then generates highly optimized code targeting multiple hardware architectures such as multicores, GPUs, FPGAs and distributed clusters. The goal of this project is to integrate Tiramisu within TensorFlow and use it to generate faster DNNs.

We have already integrated Tiramisu with success in two DSL compilers: Halide and Julia. Tiramisu extended Halide and Julia with many new capabilities including the ability to express new algorithms, the ability to perform new complex loop nest transformations and the ability to generate efficient code for more architectures. It led to speedups reaching 4x in Halide and 16x in Julia due to optimizations enabled by Tiramisu.

Preliminary experiments in implemented DNN convolutions in Tiramisu showed that convolutions generated by Tiramisu are 1.5 times faster than convolutions implemented in the Intel MKL-DNN library (the most optimized DNN library for Intel architectures).

Requirements
Knowledge of TensorFlow and deep learning.
Knowledge of compilers.
Taking 6.172/6.871 (Performance Engineering of Software Systems) is plus.
Faster DNNs in TensorFlow
"Parallel Streaming Graph Analytics"
Faculty Advisor: Julian Shun
Mentor(s):
Contact e-mail: jshun@mit.edu
Research Area(s): Computer Systems, Theoretical Computer Science
Graph algorithms have applications in a variety of domains, such as social network and Web analysis, computational biology, and machine learning. Analyzing large graphs quickly requires designing high-performance parallel algorithms. Furthermore, graphs are changing quickly, requiring data structures that allow efficient parallel updates and dynamic algorithms that respond to changes without recomputing from scratch. This project involves the design and implementation of parallel dynamic graph algorithms and possibly a programming framework that improves the programmability of these algorithms. This project is suitable for students who are interested in algorithm analysis and performance engineering.

Prerequisites: Performed well in 6.172 and 6.046 or have comparable experience. Committed to at least 15 hours/week.
"Theory and Practice of Parallel Algorithms"
Faculty Advisor: Julian Shun
Mentor(s):
Contact e-mail: jshun@mit.edu
Research Area(s): Computer Systems, Theoretical Computer Science
This project involves the design, analysis, and implementation of parallel algorithms for various fundamental problems in computing such as static and dynamic graph analytics, clustering, text analytics, sorting, data compression, etc. The goal is to design parallel algorithms that perform well both in theory and in practice. This project is suitable for students who are interested in algorithm analysis and performance engineering.

Prerequisites: Performed well in 6.172 and 6.046 or have comparable experience. Committed to at least 15 hours/week.
Theory and Practice of Parallel Algorithms

Total: 78

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