REU SITE - Intelligent Systems in Electrical and Computer Engineering

The goal of Prof. Furse’s project is to develop a wearable microwave imaging system for breast cancer detection. We are using spread spectrum time domain reflectometry (SSTDR), which is a new measurement modality for reflections and transmissions that is much faster (seconds rather than minutes), smaller and less expensive than traditional measurements made with a vector network analyzer (VNA).

We will be working on several aspects of this project including: (1) antenna design for multi-ultrawide band tests, (2) optimizing imaging algorithms (using both simulated and measured data), (3) use of AI for detection potentially without creating an image at all, (4) extension of this idea to veterinary (equine) applications, to provide an early-stage commercial option, (5) evaluation of existing and future hardware options. The desired outcome of this project are methods and assessments of reflection/transmission measurement and software for detecting changes in the breast.

Our recent papers.

Our work aims to understand what can be observed in challenging environments such as fog, snowstorms, sandstorms, and low-light conditions. We investigate how properties of the electromagnetic field—like polarization, coherence, and spectrum—can reveal additional scene information. This leads to questions about co-optimizing photonic hardware and software, including machine learning, to enhance observation and detection.

Through simulations and experiments, we explore both scientific questions and practical applications, such as improving autonomous vehicle performance across land, sea, air, and space, and investigating biological processes like temperature effects on neural activity. We also develop novel sensors for spectro-polarimetric astronomical  imaging.

Our lab encourages cross-disciplinary collaboration and has produced multiple publications led by REU students, including work on machine-learning-enabled defogging1 and predicting thermal images from RGB sensors.2

The student will run simulations and perform experiments to evaluate optical systems under various lighting conditions. If time allows, outdoor measurements will be conducted. The student will have access to a supercomputing cluster (TACC) and a fully equipped optics lab. For more details on recent publications, visit https://www.rajeshmenon.net

The goal of this project is to enhance the capabilities of neural networks to function robustly and flexibly in real-world settings with a focus on machine vision applications. Recent advances in machine learning and hardware have created artificial neural networks (ANNs) that have achieved high accuracies in some applications, e.g., image classification. In spite of these advances, such systems cannot match the performance, robustness, flexibility, or energy efficiency of the human brain. It has been hypothesized that this gap is due to the powerful properties of the brain as a computing machine. To bridge this gap, recent research has created simplified models of biological neurons and used them to create spiking neural networks (SNNs) [1, 2].

Advancing the computational capabilities of SNNs necessitates new signal encoding strategies and learning algorithms that can incorporate the different brain’s computational and circuit principles, such as high parallelism, compression, mixed digital and analog signal processing and communication, nonlinearity, or stochasticity. The students will investigate machine vision applications, which exploit the properties of neuro-inspired SNNs and perform basic comparison studies of these SNNs and other ANN systems for artificial vision applications. Example questions to be answered include: (1) What are the architectural and algorithmic bottlenecks in existing SNNs for energy efficiency, and computational robustness and flexibility? (2) How does incorporating the brain inspired temporal coding paradigm impact the real-time processing capabilities of the new SNNs?  The desired outcome of this project is SNN solutions for machine vision, with superior real-time processing and learning capabilities of ANNs by exploiting the benefits of brain-like temporal codes and mixed digital and analog processing. The student are expected to have basic knowledge of machine learning and some experience of how to train simple neural network architectures.

[1]        B. V. Benjamin, Peiran Gao, Emmett McQuinn, Swadesh Choudhary, Anand R. Chandrasekaran, Jean-Marie Bussat, Rodrigo Alvarez-Icaza, John V. Arthur, Paul A. Merolla, and Kwabena Boahen, “Neurogrid: A mixed-analog-digital multichip system for large-scale neural simulations,” Proceedings of the IEEE, vol. 102, no. 5, pp. 699-716, 2014.

[2]        L. Khacef, Nassim Abderrahmane, and Benoit Miramond, “Confronting machine-learning with neuroscience for neuromorphic architectures design,” 2018 International Joint Conference on Neural Networks (IJCNN), pp. 1-8, 2018.

The U.S. Navy is working to support the Marine Corps by providing robust communication capabilities over land and along coastlines.  This project will involve generating finite-difference time-domain (FDTD) simulations of ultra-high frequency (UHF) radio frequency signal propagation in realistic 3-D environments.  There are currently many unknowns about how radio waves interact and reflect off of vegetation, like trees and shrubs, etc., and current prediction techniques ignore many 3-D geometries and features.  The student(s) working on this project will first learn the FDTD method through hands-on activities developed by Prof. Simpson via online lectures and regular meetings to answer questions, etc.  While working through the online material, the student(s) will generate 1-D, 2-D, and then 3-D FDTD models, and they will apply them to a variety of applications, including remote sensing of people buried by avalanches (to help find victims when they are not wearing a beacon) and solving an electromagnetic interference problem at a construction site that is causing burns and shocks to the workers.  Next, the student(s) will create and apply FDTD models with the goal of generating accurate UHF signal interactions with complex, realistic 3-D environments to support the Navy project.  This work will involve machine learning methods and possibly stochastic FDTD.  The FDTD method is a very robust method that is helpful to list on your resume when applying for jobs at Apple, Intel, Amazon, COMSOL, RemCom, National Labs, etc.  Please reach out to Prof. Simpson with any questions.

This REU project aims to investigate spatial light modulators operating in the terahertz frequency range for implementation in diffractive optical neural networks. By leveraging the high bandwidth and parallelism of such, the project explores how tunable SLMs can enable real-time optical computations that emulate artificial neural network operations. The research integrates materials, photonic devices, and AI-based design to enhance the system functionality. Ultimately, the project seeks to advance energy-efficient, high-speed hardware for next-generation artificial intelligence and optical computing systems.

List of Projects:

• Automated Neural Network Emulator
• Edge-based Machine Learning
• Side-Channel Attack Modeling
• Analog Design Methodology
• Measuring Performance of a Test Chip
• Automated Software for NN Analysis
• ML based Coil Design
• Custom DSP Algorithms for ADC Characterization

The goal of Prof. George’s project is to provide amputees with intuitive control of, and natural sensory feedback from, dexterous multi-articulate bionic arms. State-of-the-art upper-limb prostheses have become capable of mimicking many of the movements and grip patterns of endogenous human hands. Although these devices have the capabilities to replace much of the motor function lost after hand amputation, the methods for controlling and receiving feedback from these prosthetic limbs are still primitive. In the United States alone, 1 in every 200 individuals suffer from limb loss [1], and up to 50% of amputees abandon their prostheses due to ineffective control and/or a lack of feedback [2]. Utah Slanted Electrode Arrays implanted into the residual arm nerves can provide bi-directional communication between the human user and the bionic arm. The challenge is developing user-specific algorithms for reading/writing electrical information to/from the nervous system that maximize dexterity.

To achieve the goals of this project, better algorithms and more precise training data for the user-specific algorithms are needed. As such, the student will learn how to develop machine learning models using Python and MATLAB, and they will use those models to analyze electrophysiological data. They will also learn about underlying assumptions made when labelling training data, and how to minimize errors in training data.  Example questions to be answered by the project include: (1) What features of neural data contain the most predictive power for kinematic motion? (2) What role does the temporal pattern of neural action potentials play in conveying contact/texture information?  The desired outcome of this project is a list of the key components of neural data that encode sensorimotor dexterity, such that future neuroprostheses could leverage these components to enhance prosthesis adoption and patient outcomes.

[1]       E. A. B. a. T. T. Chau, “Upper limb prosthesis use and abandonment: A survey of the last 25 years,” Prosthet. Orthot. Int., vol. 31, 3, pp. 236-257, 2007, doi: doi: 10.1080/03093640600994581.

[2]       E. B. a. T. Chau, “Upper-limb prosthetics: critical factors in device abandonment,” Am. J. Phys. Med. Rehabil, vol. 86, no. 12, pp. 977-987, 2007, doi: doi: 10.1097/PHM.0b013e3181587f6c.