CSE REU Projects for 2022

Algorithmic Procedural Fairness

Faculty: Yevgeniy Vorobeychik

Algorithmic approaches are now abundant, and increasingly the automated decisions they make impact real lives. How do we ensure that such decisions are fair? Conventional approaches to algorithmic fairness focus primarily on the distribution of outcomes. However, research in procedural fairness suggests that procedural aspects, such as treating people with dignity and giving them a voice in the process, are often perceived as more fundamentally important by individuals actually affected. This project will involve a human subject study of aspects of procedural fairness in the context of algorithmic decision making. We will investigate, in particular, human perceptions of which information (features) are fair to use under what conditions, as well as the impact of the ability to contest a decision influences perceptions of fairness.

Skills Required: Strong programming skills, experience in web programming, and solid understanding of statistics and data analysis. Prior experience with Amazon Mechanical Turk is a plus.

Robust Reinforcement Learning

Faculty: Yevgeniy Vorobeychik

Reinforcement learning has seen major advances in recent years, particularly as we embedded deep neural networks into RL algorithms. However, as the space of states and actions becomes large, deep RL methods often yield models that are vulnerable to adversarial perturbations. The goal of this project will be to systematically explore the vulnerability of deep RL methods to adversarial tampering, and the efficacy of techniques for making RL robust to attack, particularly when facing attacks that can be physically realized.

Skills Required: Strong programming and mathematical background. Knowledge of machine learning foundations, both conceptual and practical. Significant experience with python, including ML libraries in python. Prior experience with deep learning and/or reinforcement learning a strong plus.

Machine Learning for Computational Imaging

Faculty: Ulugbek Kamilov

Computational imaging often deals with the problem of forming images free of artifacts and noise. REU students will work on advanced algorithms for image restoration that are based on integration of optimization and machine learning. We have developed a family of such techniques that use learned information, such as natural image features, to generate clean images from the corrupt ones. REU students will have an opportunity to learn about real-world problems in biomedical imaging, study cutting edge imaging technology, and contribute to this exciting research area.

Skills Required: Familiarity with image processing and machine learning. Knowledge of Python and/or Matlab.

Optimizing traffic flows between data centers

Faculty: Roch Guerin

The project is sponsored by NSF and in collaboration with Google and explores approaches to make the (private) networks connecting large data centers more efficient, where efficiency is measured in terms of the amount of bandwidth required to meet performance guarantees (end-to-end deadlines) of the traffic flows between those data centers. Of interest is the extent to which “reshaping” traffic before it enters the network can produce better outcomes.

Initial results have identified effective reshaping strategies for flows with deterministic traffic envelopes and performance guarantees in the form of hard end-to-end deadlines. The strategies call for sophisticated scheduling policies in the network, but are effective when worst-case guarantees are needed. For practical purposes, it is, however, of interest to evaluate if those benefits are still present when considering statistical performance guarantees and non-deterministic traffic envelopes, as well as less powerful network schedulers. As these extensions are typically analytically intractable, simulations will be used to investigate them in different scenarios (network topologies).

The project will involve working with a PhD student towards developing a number of simulation models using the ns-3 simulator. The simulations will be aimed at evaluating the performance of the proposed reshaping strategies for different traffic combinations and network topologies, and analyzing the results to formulate guidelines that network engineers can rely upon. If time allows, the investigation may extend to exploring the implementation of efficient and scalable reshaping mechanisms in end-systems. This aspect of the work would involve exploring the possible use of existing mechanisms in the Linux kernel, e.g., QDisc, and evaluating their potential suitability.

Skills Required: Familiarity with C++ and python (ns-3 uses both). Prior experience with ns-3 is useful but not required. However, some understanding of simulations is expected. Exposure to network protocols and performance modeling is also useful, though again not mandatory.

Uncovering the “Hidden Half” of plants

Faculty: Tao Ju

Roots, the “hidden half” of a plant, play many important roles including physical support of the plant, uptake of water and nutrients, and stabilization of the soil. Their functions, as well as their amazingly complex structures, have intrigued biologists for centuries. With advanced imaging technique like CT and MRI, biologists are finally able to “see” these underground forms in 3D. However, computational methods are needed to extract relevant information from the images, such as identifying root branches, measuring their length and shape, understanding their organization and architecture, and analyzing them over time.

As part of a three-institution collaboration, the REU will join an interdisciplinary team of computer scientists, mathematicians, and biologists to build automated algorithms and interactive graphical tools for image-based analysis of plant roots.

Skills Required: Experience with C++ and Python, familiarity with OpenGL, good foundation in algorithms and data structures (particularly those related to graphs).

Privacy-preserving Medical Computation

Faculty: Ning Zhang

Recent advances in genome sequencing technology have made it possible to generate highly detailed genotypes inexpensively. The ubiquitous access to genomic information has led us towards a new era of medical research and personalized medicine. On the other hand, privacy is a major concern towards broader sharing of sensitive medical information. Given the current global challenges on healthcare, open access for medical discovery is a pressing issue with societal importance.

Skills Required: C++, basic security concepts

Fault-tolerant scheduling using cover-free families

Faculty: Kunal Agrawal and Netanel Raviv

Shared communication mediums are widely used in myriad systems, such as shared bus, wireless sensor networks, and more. In order to prevent transmission errors in such mediums, it is necessary to devise a scheduling algorithm, which determines which participants send their information in which time slot. Designing such static-scheduling algorithms, however, is a complex task which involves a combinatorial notion called cover-free families. A cover-free family is a family of subsets such that no member of the family is contained in a union of two other members. The study of cover-free families is a rich an ongoing one, which dates back to works by Paul Erdos. In this project we will explore the implementation of cover-free families to design static scheduling algorithms for shared communication mediums.

Skills Required: An ideal applicant for this project will have strong mathematical maturity, especially in combinatorics and discrete mathematics, as well as passion for real-world applications.

Machine Learning with Humans in the Loop

Faculty: Chien-Ju Ho

Most machine learning models are trained from data prepared by humans.
Traditional machine learning algorithms assume data is i.i.d. drawn.
However, humans might inherit various biases in data generation, and
ignoring the bias could lead to suboptimal machine learning comes. In
this project, REU students will work on problems focusing on the human
components in machine learning, such as incorporating human behavior
models in machine learning frameworks and exploring how human behavior
impacts the design of machine learning algorithms.

Skills Required: A strong mathematical background and proficiency in programming.

High Performance Computing to Benefit Science

Faculty: Roger Chamberlain, Jeremy Buhler, Ron Cytron

Many scientific disciplines benefit from advances in computational techniques. Scientific instruments can see measured signals more effectively, analysis can benefit from substantially larger volumes of data, and ultimately we can ask and answer deeper questions about the world around us.

In this project we work with experts in multiple domains of science (e.g., astrophysics, biology, nuclear chemistry) and seek to improve the performance and power efficiency of computations that benefit scientific inquiry. This often includes the exploitation of non-traditional computer architectures used to accelerate an application.

Skills Required: Programming ability; C/C++ experience a plus but not strictly required