The proposed research projects are subject to change, contingent on new mentors and/or the research interests of the students.
The use of machine learning has become ubiquitous in various areas, including hiring, lending, and risk predictions, where unbiased decision-making is of utmost importance. Many approaches have been proposed to tackle bias in machine learning models to ensure fairness in these applications. This project's primary objective is to explore the application of constrained optimization models and algorithms to enforce fairness and reduce bias in machine learning models. These models incorporate explicit constraints in the training optimization problem to ensure fairness based on specific criteria, such as demographic parity or equal opportunity. We will explore various constrained optimization algorithms to train these machine learning models and evaluate their performance using publicly available datasets. Prerequisites: Python or Matlab (preferred python); Basic Linear Algebra & Probability. Mentor: Raghu Bollapragada, Assistant Professor, Graduate Program in Operations Research and Industrial Engineering.
With the increasing demand for new, cleaner chemical processes, the use of living systems (like microorganisms) has become of increased interest. One approach is towards engineering microorganisms, such as bacteria and yeast, as cellular factories to produce chemicals of interest. Natural, internal regulators of these microorganisms called small RNAs (sRNAs) are promising tools for the control of these organisms for engineering purposes. However, it is largely unknown where these sRNAs are encoded in the cellular genome and which of these sRNAs are actually useful for the given engineering goal. The Chemical Engineering group is developing a bioinformatics pipeline including RNA-sequencing analysis to identify sRNAs and rank them according to their probability of impact as regulators for engineering goals. The REU student will help automate this process, adapt it across different organisms, combining multiple existing tools (R, Python, web) and manual processes into a more user-friendly package. Prerequisites: Basic programming in any language. Learning outcomes: (1) understand how to parse large scale biological data for analysis, (2) work in an interdisciplinary environment with biologists and molecular engineers to devise useful algorithms that can be user-friendly, (3) troubleshoot programming skills in a completely applied setting, (4) educate others (without a strong programming background) to be able to annotate and run algorithms designed and created. Mentors: Lydia Contreras, Associate Professor, Miller Endowed Faculty Fellowship in Chemical Engineering.
Digital Rocks is a data portal for fast storage and retrieval, sharing, organization, and analysis of images of varied porous micro-structures. It has the purpose of enhancing research resources for modeling/prediction of porous material properties in the fields of Petroleum, Civil and Environmental Engineering as well as Geology. This platform allows managing, preserving, visualization, and basic analysis of available images of porous materials and experiments performed on them, and any accompanying measurements (porosity, capillary pressure, permeability, electrical, NMR and elastic properties, etc.) required for both validations on modeling approaches and the upscaling and building of larger (hydro)geological models. Prerequisites: proficient in Python and willing to learn 2D and 3D visualization of complex materials. Learning outcomes: (1) organize and preserve images and related experimental measurements of different porous materials, (2) improve access to these images for a wider community of geosciences and engineering researchers not necessarily trained in CS or data analysis, (3) enhance productivity and enable scientific inquiry and engineering decisions. Mentors: Maša Prodanović, Associate Professor, Chevron Centennial Teaching Fellow in Petroleum Engineering.
The REU student will investigate utilizing publicly available datasets with immersive and/or interactive visualization techniques to create an interface between the technology and the scientific or societal problem at hand. An emphasis is placed on the selection and creation of the appropriate visualization modality (e.g. 2D or interactive vs. VR immersive) for effective exploration of the research question. This project allows the opportunity for motivated students to select a research problem based on a compelling environmental or societal problem that is of interest to them, and work through the process of evaluating the available datasets, performing statistical analysis, and designing, implementing, and evaluating a visualization. There are also several projects underway that the student can select from that use environmental, crime, or poverty/homelessness data sets. One example of an established project is an interactive virtual reality environment “A Walk along Waller Creek” done in collaboration with Dr. Mary Poteet (UT Jackson School of Geosciences) that shows correlations between water quality and fish biodiversity using a local Austin watershed Waller Creek. The REU student will access geolocated datasets concerning collected water quality sensor data and biospecimens as cataloged by the iNaturalist database. The student will investigate correlations between physical properties of water quality and aquatic life/biodiversity (using GIS, R, and python tools) and represent this correlation using immersive or interactive technologies. Learning Outcomes: (1) statistical analysis and representation of data, (2) design and development of an immersive/interactive visualization, (3) utilizing game engine technology for scientific research and education. Mentor: Anne Bowen, Research Associate, Data Visualization
Stroke is the leading cause of disability in the industrialized world, and approximately 50% of stroke survivors will suffer long-term motor impairments severe enough to cause disability or require the assistance of others for basic daily tasks. Researchers are investigating various strategies to help enhance recovery following stroke. Optogenetics is a powerful technology that allows for the control of specific neurons with high temporal resolution and has shown promising potential therapeutic use in other neurological disorders and diseases such as epilepsy and Parkinson’s disease; however, its potential therapeutic use in stroke has not been thoroughly investigated. This project aims to do so, specifically looking at the effects of optogenetic stimulation in stroke mice. The REU student will program the optogenetic stimulation, using Arduino, that will be delivered during the experiments, as well as use multiple software for analysis of animal behavior. These include Bonsai, a software that allows for automated online tracking of animal behavior as well as use DeepLabCut, a toolbox used to compute 3D pose estimates. Prerequisites: Python experience. Learning Outcomes: (1) design and development of computer vision models, (2) development of automated data pipelines and deployment on HPC hardware. Mentor: Huiliang (Evan) Wang, Assistant Professor, Biomedical Engineering.
Women’s health represents one of the most pressing health policy issues nationally. In no medical specialty are the deficiencies of medical evidence more pronounced than in women’s health, especially in obstetrics. Over the course of the human lifespan, birth is one of the most dangerous health episodes for both mother and baby. Worldwide, between 2.6 and 4 million pregnancies result in stillbirth annually. If stillbirths were included among the causes of human mortality, they would rank as the third leading cause of death following ischemic heart disease and stroke. Additionally, the U.S. is experiencing a health crisis resulting from decades of increasing health care expenses associated with ever-smaller improvements in health outcomes, and the development of an efficient, modern healthcare system has become an increasingly critical concern. The U.S. spends five times more per capita on health care than countries with similar life expectancies, and it has become increasingly clear that the rising number of tests and interventions do not improve health unilaterally - 5% of the population accounts for 51% of total healthcare expenditures. This project will support research in computational health, primarily researching and developing methods to support individualized medicine and risk visualization. Prerequisites: science or engineering major with interest in developing computational and data analytics skills. Learning Outcomes: (1) development and analysis of statistical and machine learning algorithms for individualized medicine, (2) implementation of visualization methods to communicate risk, (3) development of data analysis and mining methods. Mentors: Kelly Gaither, Associate Professor, Women’s Health, UT Dell Medical School; Director of Health Analytics, TACC, Karl W. Schulz, Associate Professor, Women’s Health, UT Dell Medical School; Research Associate, Oden Institute for Computational Engineering and Sciences, and Justin Drake, Research Associate, Health Analytics, TACC and Assistant Professor, Women’s Health, UT Dell Medical School.
This is a five-year, multi-institution, NSF-funded project to forward-model earthquakes in subduction zones across a range of Spatio-temporal scales using HPC, and to integrate data from diverse sensor networks for validation and uncertainty quantification. In particular, surface deformation from geodesy, deep deformation as indicated by small earthquakes, and seismic wave speeds are observed to change before, during, and after an earthquake. The project seeks to integrate those observations, represent them in 4-D visualizations, and merge them with a range of forward modeling results that seek to capture these processes. The exploratory part is to then define workflows for quantitative geospatial analysis, in particular comparisons between observations and synthetics. Prerequisites: Basic familiarity with Python. Learning outcomes: (1) Process heterogeneous data, convert to a common visualization format, explore and display different data layers, (2) define new metrics for analysis of correlation and explore different representations of information, (3) enhance scientific discovery for earthquake mechanics and define knowledge gaps. Mentor: Thorsten Becker, Shell Foundation Distinguished Chair in Geophysics, Jackson School of Geosciences.
