Friday, November 06, 2026
10:20 – 11:10 a.m. in ETB 1020
Dr. Yang Shen
Professor
Department of Electrical and Computer Engineering
Title: “Learning the Geometry and Language of Biomolecules: From Generation and Perturbation to Controllable Design”
Talking Points:
* Coupled discrete and geometric diffusion enables joint generation of protein sequences and structures.
* Physics-informed generative dynamics model how protein structures respond to chemical perturbations.
* Uncertainty-aware Pareto preference alignment steers protein design across competing functional objectives.
Abstract:
Proteins, the molecular workhorses of life, can be represented simultaneously as one-dimensional amino-acid sequences, three-dimensional geometries, and natural-language descriptions of their functions. Learning across these representations creates distinctive challenges for artificial intelligence: biomolecular states combine categorical amino-acid identities with continuous geometric variables, protein structures and functions change under genetic and chemical perturbations, and useful designs must balance competing properties measured or predicted with uncertainty.
This talk presents multimodal generative methods for understanding and designing proteins. JointDiff couples discrete and geometric diffusion to generate protein sequences and structures jointly while incorporating computational or experimental feedback. A physics-informed Schrödinger bridge extends static generation to ligand-induced structural responses by modeling transitions between unperturbed and chemically perturbed protein geometries. ST-PARM enables inference-time steering of protein language models across competing objectives, such as activity, stability, and solubility, through uncertainty-aware preference alignment for broad Pareto coverage. I will also briefly discuss how multimodal protein language models connect sequence and structure with functional descriptions, and why predictive accuracy alone may not ensure biologically grounded explanations.
Together, these studies illustrate a progression from representing biomolecular states to modeling their responses to perturbations and steering designs toward desired functional trade-offs. I will conclude with a vision for biomolecular world models that learn how molecular states respond to interventions and connect those responses across scales to cellular states and organismal phenotypes, providing building blocks for mechanistic understanding of disease and the design of therapeutic interventions.
Biography:
Dr. Yang Shen is Professor of Electrical and Computer Engineering at Texas A&M University, with affiliated appointments in Computer Science and Engineering, Translational Medical Sciences, and the Institute of Biosciences and Technology. He received his Ph.D. in Systems Engineering from Boston University and completed postdoctoral training in Biological Engineering and Computer Science at MIT. He develops machine-learning and optimization methods for understanding, predicting, and designing biological systems, spanning graph and multimodal learning, geometric and generative modeling, uncertainty quantification, domain adaptation, and physics-informed AI. His applications connect protein sequence, structure, function, interactions, and responses to genetic and chemical perturbations toward generalizable, explainable, and actionable biological AI. He has received two NIH Maximizing Investigators’ Research Awards and an NSF CAREER Award.
Please join us on Friday, 11/06/26 at 10:20 a.m. in ETB 1020!
Dr. Yang Shen’s homepage: https://engineering.tamu.edu/electrical/profiles/shen-yang.html/
]]>
Friday, November 20, 2026
10:20 – 11:10 a.m. in ETB 1020
Zhiwen Fan
Assistant Professor
Department of Electrical and Computer Engineering
Title: “From Human Data to Robot Skills: Scaling Vision, Touch, and Behavior”
Talking Points:
* Physics-grounded 3D world models for realistic hand-object interactions
* Scaling human vision-and-touch data for dexterous robot learning
* Safer robot behaviors through policy editing and failure recovery
Abstract:
Robotic manipulation remains limited by the lack of scalable training data that captures physical interactions, particularly contact and force. Human activities provide abundant demonstrations, but transferring them to robots remains challenging.
In this talk, I will present our recent research in three directions: (1) 3D- and physics-grounded world models for realistic hand-object interactions; (2) TouchScale, a 500-hour human vision-and-touch dataset, and approaches to scaling tactile learning and transferring human skills to robots; and (3) methods for learning safer robot behaviors and recovering from failures. Together, these efforts aim to enable scalable, reliable robot manipulation, with laboratory automation as a motivating application.
Biography:
Zhiwen Fan is an Assistant Professor of Electrical and Computer Engineering and affiliated faculty in Mechanical Engineering at Texas A&M University, where he leads the Physical Automation Intelligence (PHAI) Lab. He received his Ph.D. from UT Austin in 2025. His research focuses on machine learning, robot learning and biomedical AI. His honors and research support include the Meta AI Impact Award (2026), NVIDIA Academic Grant Program (2026), Qualcomm Innovation Fellowship (2022), and Best Paper Awards at CVPR and ACM Multimedia workshops (2025).
Please join us on Friday, 11/20/26 at 10:20 a.m. in ETB 1020!
Zhiwen Fan’s Github homepage: https://phai-lab.github.io/
]]>
Friday, October 23, 2026
10:20 – 11:10 a.m. in ETB 1020
Su-in Yi
Assistant Professor
Department of Electrical and Computer Engineering
Title: “Post-BackProp Deep Learning Algorithms for Analog In-Memory Computing using Memristor Crossbars”
Talking Points:
*Shifting memristors from AI inference to training boosts energy efficiency
*Traditional Backpropagation lacks hardware precision and differs from brain function
*Introducing “post-BackProp” algorithms tested on scalable hardware up to 300-billion synapses
Abstract:
Memristors (or non-volatile memories) crossbar-based analog in-memory computing for artificial intelligence (AI) accelerations emerged in the recent decade to achieve green and sustainable AI through extreme energy efficiencies. Mostly, the inference of artificial neural networks stemming from inherently repetitive vector-matrix-multiplications (VMMs) has been focused so far in the community. Now it is timely to expand the capability of memristor array hardware further into the training, which possesses further requirements such as high precision and large memory. Unfortunately, the conventional training algorithm, backpropagation (BackProp or BP), is not compatible with most memristor array hardware due to low precisions, stemming from stochasticity in their conductance values. Moreover, BackProp itself has been almost 40 years since the invention by Hinton, a Nobel Laureate in Physics in 2024, and has room for improvement in biological plausibility to ultimately mimic how the brain functions with enormous power efficiencies compared to modern computers. In this talk, post-BackProp algorithms for high energy efficiencies together with memristive analog hardware such as resistive random-access memory (ReRAM) with 4000 artificial synapses and 3D NAND flash memory with 300 billion artificial synapses will be presented. These biologically plausible deep learning algorithms include memristor activity difference energy minimization (MADEM), forward target propagation (FTP), and memristive predictive coding (PC).
Biography:
Dr. Su-in Yi received the M.S. degree from the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea, in 2012, and the Ph.D. degree from Texas A&M University, College Station, TX, USA, in 2017. Before joining the faculty, he was a Staff Engineer with Semiconductor Research and Development Center at Samsung Electronics for 4 years and a Research Engineering at Hyundai Wia for 1 year. He is currently an Assistant Professor and Raghavendra K. Pandey Faculty Fellow with the Department of Electrical and Computer Engineering, Texas A&M University. His research interests include neuromorphic computing devices and computer architecture, non-volatile memory (NVM) reliability and security, and hardware-compatible AI learning algorithms. He was the recipient of the US AirForce Office of Scientific Research (AFOSR) Young Investigator Program (YIP) Award in 2025.
Please join us on Friday, 10/23/26 at 10:20 a.m. in ETB 1020!
Su-in Yi’s google scholar profile: https://engineering.tamu.edu/electrical/profiles/yi-suin.html
]]>
Friday, October 16, 2026
10:20 – 11:10 a.m. in ETB 1020
Thai Duong
Postdoctoral Researcher in Computer Science
Rice University
Title: “Exploiting Physics Structures in Dynamics Learning, Planning and Control for Efficient Robot Autonomy”
Talking Points:
* Preserving physics structures in machine learning models enables efficient dynamics learning with accurate long-term predictions
* Preserved physics structures facilitate control designs by forming an important bridge between control theory and machine learning
* Physics structures unlock fast planning of robot motions for reactive robot operation, e.g., in changing environments.
Abstract:
Recent advances in robot autonomy (e.g., motion planning and control) have made robot systems more performant and reliable by relying on analytical models of the systems. Meanwhile, machine learning has equipped robots with the ability to acquire complex models and skills, usually from a large amount of data. While the real world is rich in (often simplified) analytical models for traditional robot autonomy, it is so messy and complex that learning is crucial to compensate. Therefore, a major challenge in robotics today is to come up with a new paradigm that embraces the advances from both model-based techniques and learning-based methods for long-term, reliable and intelligent robot systems. In this talk, I will present an approach to unify machine learning, robot planning, and control via known physics structures, e.g., universal physics laws, for effective robot autonomy. In the first half, I will discuss how preserving physics structures in machine learning models enables efficient learning of robot dynamics and facilitates control designs for accurate trajectory tracking using the learned models. In the second half, I will talk about how physics structures unlock extremely fast and reactive planning of robot trajectories under dynamics constraints. My techniques lay the groundwork for self-improving long-horizon robot autonomy via a virtuous learning-planning-control-data circle. Extensive validation of my techniques on different robot platforms, e.g., mobile robots, legged robots, and manipulators, will also be presented.
Biography:
Thai Duong is currently a postdoctoral researcher in Computer Science at Rice University, Houston, TX, USA, working with Prof. Lydia Kavraki. He obtained his Ph.D. degree in Electrical and Computer Engineering from the University of California, San Diego in 2024, under the direction of Prof. Nikolay Atanasov. His research focuses on developing efficient, safe and reliable autonomous robot systems by integrating robot learning, planning and control in a principled manner. His work was recognized by the 2026 Robotics: Science and Systems (RSS) Pioneers program, the 2025 Best Paper Award by the IEEE RAS Technical Committee on Robot Control, and an Honorable Mention of IEEE Transactions on Robotics King-Sun Fu Memorial Best Paper Award in 2025.
Please join us on Friday, 10/16/26 at 10:20 a.m. in ETB 1020!
Thai Duong’s personal webpage:: https://thaipduong.github.io/ /
Host : Varun Murali
Our students showcase their research, scholarly, and creative activities!
Friday, Sept. 25, 2026 · 1:30 – 3:30 PM
Wisenbaker Engineering Building (WEB) Lobby
| Student Names | Poster Title | Faculty Advisor |
|---|---|---|
| Arjuman Ara Mimi | Combinatorial Optimization using TRNG from 3D NAND Flash | Dr. Suin Yi |
| Chia-Hsiang Shyu | Distributed Algorithm for Multi-Hop Linear Contextual Bandit | Dr. I-Hong Hou |
| Daniel Puckett | Bottleneck Interactions on Out-of-Order CPUs | Dr. Paul Gratz |
| Dayou Li | EgoHOI: Egocentric World Model for Photorealistic Hand-Object Interaction Synthesis | Dr. Zhiwen Fan |
| Fatemeh Doudi Visnu Teja Kunde Mahdi Farahbakhsh | Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages | Dr. Dileep Kalathil |
| Fenghua Wu | Spec2Assertion | Dr. Jiang Hu |
| Haodong Chang | Fast Cross-Operator Optimization of Attention Dataflow on Spatial Accelerators | Dr. Jiang Hu |
| Jinzhan Li | True Random Number Generation Using 3D NAND Flash | Dr. Suin Yi |
| Joshua Mashburn | Targeted Wearout Attacks in Microprocessor Cores | Dr. Paul Gratz |
| Lei Wang | R-Max: Extending Bélády’s MIN with Prefetching to Bound Realistic Cache Performance | Dr. Paul Gratz |
| Luis Loo | An Agentic AI Science Community for Automated Neural Operator Discovery | Dr. Ulisses Braga-Neto |
| Mincheol Seong | Fast Constraint Adaptation in Constrained Reinforcement Learning via Zero-Shot Primal-Dual Updates | Dr. I-Hong Hou |
| Sabyasachi Gupta | Beyond Abstraction Overhead: Cross-Layer Characterization of Edge Serverless Execution | Dr. John Lusher II |
| Saichand Samudrala | Performance Optimization for 3DGS Inference | Dr. Paul Gratz |
| Satota Mandal | LASER: LLM-Assisted Side-channel Leakage Elimination and Resistance in Assembly Code | Dr. JV Rajendran |
| Seyed Ali Ghazi Asgar | Quietprint | Dr. Narasimha Reddy |
| Yiming Gao | PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects | Dr. Yang Shen |
| Yuxuan Liu | Protein Structure Perturbation Modelling with Ligand | Dr. Yang Shen |
Friday, September 25, 202610:20 – 11:10 a.m. in ETB 1020
Marko Scrbak
Senior Member of Technical Staff at AMD
Title: “Power-Aware Performance in the Era of Extreme-Scale Computing”
Talking Points:
* Power management is a critical efficiency lever and stretches beyond DVFS
* Power modelling lacks serious details for effective co-design
* Cross-stack power optimizations are necessary for bridging the efficiency gap
Abstract:
As we push the boundaries of extreme-scale computing in the AI era, power and energy have become first-order constraints, fundamentally shaping system performance. This talk explores the critical interplay between power consumption and performance and discusses new research directions enabled by recent advancements in agentic AI solutions. We will uncover details that govern power management under power-limited conditions and discuss their effect on performance variability and modelling complexity. Understanding these issues is essential for guiding power efficiency research in the right direction to design and optimize AI systems under real-world energy constraints.
Biography:
Marko Scrbak joined AMD in 2018 after several successful internships during the FastForward2 and PathForward programs working on exascale computing research.His current research focuses on power efficient accelerated computing. Past research includes CPU frontend, cache, and memory systems optimization, application performance/power profiling, and near-data computing. Marko holds a Ph.D. and M.S. degree in Computer Science and Engineering from UNT and a Bachelor’s in Computing from his hometown Zagreb, Croatia. In his free time, Marko shares his passion for cooking, traditional (international) music, and soccer (football).
Please join us on Friday, 9/25/26 at 10:20 a.m. in ETB 1020!
Note: Dr. Scrbak is part of Electrical and Computer Engineering’s AMD Week and
Day 5’s: Industry Affiliates (IAP) Day which includes:
* Student Networking Event: WEB 232 9 a.m. – noon
* Graduate Poster Event 1:30 – 4:00 p.m.: WEB Lobby: Poster Info. Here
]]>
Friday, October 9, 202610:20 – 11:10 a.m. in ETB 1020
Yu Xiang
Assistant Professor, Computer Science
University of Texas at Dallas
Title: “Toward Generalizable Robot Manipulation: Cross-Embodiment Learning and Real-World Evaluation”
Abstract:
Generalization remains a central challenge in robot manipulation. In this talk, I will discuss our recent work addressing two complementary aspects of this challenge: generalization across robot embodiments and evaluation of real-world generalization. First, I will introduce the Unified Hand Action Space (UHAS), which represents dexterous hand actions through deformation of a canonical sphere, enabling policy learning and transfer across substantially different robotic hands. Second, I will present VLA-Replica, a low-cost and reproducible framework for evaluating vision-language-action (VLA) policies under both in-distribution and out-of-distribution conditions. Through evaluations of recent manipulation models, I will discuss their capabilities and limitations in real-world settings. Together, these efforts highlight the importance of both transferable action representations and rigorous real-world evaluation toward more generalizable robot manipulation.
Talking Points:
* Cross-embodiment robot manipulation with a unified action representation
* Learning manipulation policies that transfer across different robotic hands
* Reproducible real-world evaluation of vision-language-action (VLA) models
* Understanding the capabilities and limitations of generalist robot policies under real-world distribution shifts
Biography:
Yu Xiang is an Assistant Professor of Computer Science at The University of Texas at Dallas, where he leads the Intelligent Robotics and Vision Lab (IRVL). His research focuses on robotics and computer vision, with an emphasis on robot manipulation, robot learning, and generalizable robotic systems. Before joining UT Dallas, he was a Senior Research Scientist at NVIDIA and conducted postdoctoral research at Stanford University and the University of Washington. He received his Ph.D. in Electrical Engineering from the University of Michigan. He is a recipient of the Sony Research Award (2022) and the NVIDIA Academic Grant Program Award (2024). He is an IEEE Senior Member and serves as an Associate Editor for IEEE Robotics and Automation Letters (RA-L).
Please join us on Friday, 10/09/26 at 10:20 a.m. in ETB 1020!
Yu Xiang’s personal webpage:: https://yuxng.github.io/
Host : Dileep Kalathil
]]>
Friday, October 2, 202610:20 – 11:10 a.m. in ETB 1020
Debashri Roy
Assistant Professor, Department of Computer Science and Engineering at The University of Texas at Arlington
Title: “AI-Driven NextG Networks: Multimodal Fusion and ISAC for Proactive Network Management”
Talking Points:
* Multimodal sensor fusion (LiDAR, vision, GPS) for low-latency mmWave beam prediction in vehicular networks
* YOLO-based spectrogram analysis for robust radar parameter estimation and shared spectrum coexistence under heavy noise
* Multi-Agent Reinforcement Learning (MARL) and conflict graphs for proactive, cross-layer network resource orchestration
Abstract:
Next-generation (NextG) wireless networks are rapidly evolving from reactive communication systems to proactive, context-aware infrastructures. To realize this vision, this talk presents a unified, AI-driven framework that bridges Multimodal Sensor Fusion and Integrated Sensing and Communication (ISAC) to enable true proactive network management. By transforming raw physical data—spanning environmental sensors and RF waveforms—into edge-native intelligence, networks can anticipate user intent and resource needs before they emerge. We will first explore how fusing non-RF multimodal data (LiDAR, RGB/depth cameras, and GPS) overcomes the latency bottlenecks of exhaustive beam search, predicting optimal millimeter-wave (mmWave) beam pairs for highly mobile vehicular networks. Building on these predictive insights, the presentation will detail how to potentially resolve complex multi-user contention using Contextual Conflict Graphs and Multi-Agent Reinforcement Learning (MARL), dynamically orchestrating cross-layer resource provisioning across the PHY, MAC, RLC, and Application layers.
Transitioning into the ISAC domain, the talk will address the critical challenge of secure spectrum coexistence within these autonomous ecosystems. We will highlight how deep learning architectures, specifically YOLO-based spectrogram analysis, can accurately estimate radar parameters and detect anomalous interference under heavy noise, far surpassing the capabilities of classical signal processing. Because continuous ISAC data processing creates severe computational bottlenecks, we will introduce edge-native decentralized computing algorithms, such as Explainable AI (XAI)-Guided Co-Aware Partitioning. This approach intelligently distributes complex multimodal fusion models and Hierarchical Graph Neural Networks (H-GNNs) across heterogeneous edge devices, jointly optimizing for hardware capacity, communication latency, and feature importance. Finally, the talk will demonstrate how these theoretical AI models are validated through advanced emulation platforms and physical toolchains to accelerate the real-world prototyping of proactive NextG networks.
Biography:
Debashri Roy is an assistant professor at the Department of Computer Science and Engineering at The University of Texas at Arlington since Fall 2023. Prior to this, she was a research scientist at Northeastern University during 2020-2023. She received her Ph.D. and M.S. in Computer Science from the University of Central Florida in May 2020. Her research interests involve machine learning based applications in wireless communication domain, targeted to the areas of integrated sensing and communication, deep spectrum learning, millimeter wave beamforming, multimodal fusion, satellite communication, networked robots for next-generation communication. Dr. Roy’s work has been recognized with IEEE IPCCC (2018) and IEEE INFOCOM Workshop (2025) best paper awards. She was awarded multiple National Science Foundation (NSF) grants including CAREER. She serves on the editorial board for IEEE Transactions on Mobile Computing, Elsevier Adhoc Networks, Computer Networks, and Computer Communication journals. She has authored around 23 journal/magazine papers, 39 conference papers, 2 book chapters, and 4 provisional patents. She is also named as one of the 10 women globally for N2Women: Rising Stars in Networking and Communications List in 2023 and “100 Brilliant and Inspiring Women in 6G
List’” for 2026.
More Details: https://twistlab.uta.edu/about-pi/
Host: Srinivas Shakkottai
Please join us on Friday, 10/02/26 at 10:20 a.m. in ETB 1020! ]]>