Date: Wednesday, April 29, 2026, at 12:00 p.m.

Location: Coda, Room 114 (Google Maps link)

Lunch provided!

Title: Safety Alignment of Generative Foundation Models

Abstract: Modern LLMs are safety-aligned through supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF) to mitigate harmful, undesirable, or disallowed outputs. Despite ongoing progress, LLMs still exhibit critical safety gaps: models can be jailbroken into revealing harmful content, often overrefuse benign queries, and fail to maintain safety under adversarial scenarios. My dissertation research advances the safety alignment of generative foundation models by developing principled tools, architectures, and training methods that strengthen their robustness and reliability at scale. Specifically, this thesis focuses on three complementary thrusts: a) Understanding and shaping the safety landscape of LLMs, b) Internalizing safety in agentic reasoning intelligence, and c) Grounding safety and robustness in multimodal perception.

Bio: Anthony is a CS Ph.D. candidate at Georgia Tech working with Polo Chau. His thesis research has elevated foundational AI efforts at Nvidia, Meta, IBM, Intel, and ADP via internships and collaborations, and has resulted in several first-author publications and awards at NeurIPS, ACL, ICCV, EMNLP, CVPR, and BMVC. His research has contributed to the AI foundation of multiple funded industry research grants totaling over $1.4M. Learn more about him at shengyun-peng.github.io.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

Title: Recent Advances in the Design and Simulation of Approximate Multi-Species Collisional Models in Computational Kinetic Theory

Abstract: Kinetic descriptions of particle systems are characterized by a balance of phase-space advection and particles collisions. Often the operators that model collisions are the most expensive part of a kinetic simulation, and simplified models are desired. In this talk, I will present some of these models and discuss recent extensions to the multispecies setting.

Bio: Cory Hauck is an applied mathematician and group leader of the Multiscale Methods and Dynamics Group. His research interests cover various aspects of numerical analysis, computational physics, scientific computing, and mathematical tools for scientific data. His primary research area lies is the development of numerical methods for kinetic equations and related models. These models are characteristic by multiscale phenomena, and their simulation requires advanced mathematical tools in modeling and computation.

Dr. Hauck received his Ph.D. in Applied Mathematics from the University of Maryland in 2006.  Before moving to Oak Ridge, he did a postdoc in the Computational Physics and Methods Group and at the Center for Nonlinear Studies at Los Alamos. He serves on the editorial boards of SIAM Multiscale Modeling & Simulation, Journal of Scientific Computing, and others. He has received Householder Fellowship from ORNL and Early Career Award from the Department of Energy (ASCR).

Location: Klaus Atrium

Who: All students are welcome to participate. 

Register on the site or use the link provided.

Title: Breaking the Scaling Wall in Distributed Deep Learning

Abstract: Significant advances in computer architecture (development of extremely powerful server-class GPUs) and parallel computing (scalable libraries for dense and sparse linear algebra) have contributed to the on-going AI revolution. In particular, distributed training of deep neural networks (DNNs) relies on scalable matrix multiplication algorithms and efficient communication on high-speed interconnects. Pre-training and fine-tuning large language models (LLMs) with hundreds of billions to trillions of parameters and graph neural networks (GNNs) on extremely large graphs requires hundreds to tens of thousands of GPUs. However, such training often suffers from significant scaling bottlenecks such as high communication overheads and load imbalance.

In this talk, I will present several systems research directions that directly impact AI model training. First, I will describe my group's work in using a three-dimensional parallel algorithm for matrix multiplication in large-scale LLM training.  Second, I will demonstrate the application of the same algorithm to full-graph and mini-batch GNN training when working with extremely large graphs. Finally, I will also discuss the need for scalable collective communication routines for large-scale DNN training.

Bio: Abhinav Bhatele is an associate professor in the department of computer science, and director of the Parallel Software and Systems Group at the University of Maryland, College Park. His research interests are broadly in systems and AI, with a focus on parallel computing and distributed AI. He has published research in parallel programming models and runtimes, network design and simulation, applications of machine learning to parallel systems, parallel deep learning, and on analyzing/visualizing, modeling and optimizing the performance of parallel software and systems. Abhinav has received best paper awards at Euro-Par 2009, IPDPS 2013, IPDPS 2016, and PDP 2024, and a best poster award at SC 2023. He was selected as a recipient of the IEEE TCSC Award for Excellence in Scalable Computing (Early Career) in 2014, the LLNL Early and Mid-Career Recognition award in 2018, the NSF CAREER award in 2021, the IEEE TCSC Award for Excellence in Scalable Computing (Middle Career) in 2023, and the UIUC CS Early Career Academic Achievement Alumni Award in 2024.

Abhinav received a B.Tech. degree in Computer Science and Engineering from I.I.T. Kanpur, India in May 2005, and M.S. and Ph.D. degrees in Computer Science from the University of Illinois at Urbana-Champaign in 2007 and 2010 respectively. He was a post-doc and later computer scientist in the Center for Applied Scientific Computing at Lawrence Livermore National Laboratory from 2011-2019. Abhinav was an associate editor of the IEEE Transactions on Parallel and Distributed Systems (TPDS) from 2022-2024. He was one of the General Chairs of IEEE Cluster 2022, and Research Papers Chair of ISC 2023.

Date: Wednesday, April 1, 2026, at 12:00 p.m.

Location: Coda, Room 230 (Google Maps link)

Lunch provided!

Title: Fast Active-Set Thresholding Method for Nonnegative Least Squares

Abstract: Nonnegative Least Squares (NNLS) is a fundamental constrained optimization problem encountered in many applications such as image deblurring, signal processing, nonnegative matrix factorization, magnetic microscopy, and hyperspectral imaging. Active-set based methods are a common class of algorithms for solving NNLS which identify the optimal variable set of the NNLS solution. They do so by iteratively solving a series of unconstrained least squares problems, identifying which variables violate the nonnegativity constraints, and then swapping variables in/out of consideration until the optimal set of variables is found. Several variations improving upon this method exist in the literature. In this work, we propose an active-set swap heuristic which further improves upon existing active-set based methods for NNLS. Our optimizations are based upon adding multiple variables to the passive set within a threshold of the smallest gradient value and removing variables within a similar threshold of the closest boundary constraint. We leverage these optimizations to yield a Fast Active-Set Thresholding NNLS (FAST-NNLS) algorithm which significantly outperforms the existing state-of-the-art NNLS algorithms for a wide range of problems. Rigorous convergence guarantees are proven for the proposed method. We demonstrate the effectiveness of our proposed method on multiple synthetic datasets and two real-world text analysis applications. In doing so, we present the most comprehensive NNLS solver comparison in the literature to date.

Bio: Ben Cobb is a CSE Ph.D. candidate conducting research at the intersection of high-performance computing and numerical linear algebra under the co-advisement of Richard Vuduc and Haesun Park. His doctoral research pertains to various forms of tensor decompositions for data analysis and compression.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

Speaker: Steven Xia, Ph.D. Candidate, University of Illinois Urbana-Champaign

Abstract:

AI and large language models (LLMs) are ushering in a new era of software development, fundamentally reshaping how systems are built, tested, and maintained. In this talk, I will present my research on integrating modern LLM capabilities with core software engineering principles to automate complex development tasks and strengthen the coding and reasoning abilities of AI systems. I will first introduce TitanFuzz, which leverages LLMs to implicitly solve complex and heterogeneous constraints for large-scale system testing, uncovering critical bugs in real-world software, and inspiring extensive subsequent work in LLM-driven testing. Next, I will discuss Agentless, which demonstrates that a simple, lightweight framework can effectively solve challenging repository-level software engineering tasks without heavy agentic overhead. I will showcase how Agentless has been adopted by leading AI companies to evaluate their latest models, inspire new agentic frameworks, and perform post- or mid-training for agentic coding and reasoning. In addition, I will highlight my other work on automating complex software tasks and my recent work to enable runtime self-improvement in AI coding systems. Together, my work establishes a scalable and practical foundation for building more capable and reliable AI-driven software systems.

Bio:

Chunqiu Steven Xia is a PhD candidate in Computer Science at UIUC, advised by Professor Lingming Zhang. His research lies at the intersection of Software Engineering and Artificial Intelligence, with a focus on automating critical and cognitively demanding software development tasks using large language models (LLMs) and software agents. His work has significantly advanced how fundamental software problems are addressed in both academia and industry, uncovering 200+ critical bugs and vulnerabilities in widely used systems, and being adopted by leading AI companies, including OpenAI, Meta, DeepSeek, and MiniMax. His research has been recognized with an ACM SIGSOFT Distinguished Paper Award (FSE 2025), an Oral Spotlight Paper Award (COLM 2024), a CACM Research Highlight (FSE 2023), and the prestigious Amazon AICE PhD Fellowship.

Title: Diffusion Models for the Next Generation of Simulation-Based Inference: Challenges and Opportunities

Abstract: Diffusion models have recently emerged as powerful tools for simulation-based inference (SBI). SBI tackles the problem of estimating complex models from complex data, and the score-based formulation of diffusion models offers unmatched flexibility for inference and generation. This talk will synthesize recent advancements in diffusion models for SBI, focusing on key developments in training, inference, and evaluation. It will explore the potential of concepts such as guidance, score composition, flow matching, consistency models, and joint modeling. Finally, it will showcase three recent contributions from our group: 1) consistency models for near-real-time Bayesian inference; 2) compositional models for large-scale multilevel estimation, and 3) amortized Bayesian adaptive design for maximizing information gain in sequential experiments.

Bio: Stefan T. Radev is the principal investigator of the BayesOps Lab. He leads research on amortized Bayesian inference with generative AI and is a core developer and originator of the BayesFlow framework. He is particularly interested in the intersection of Bayesian modeling, deep learning, and cognitive science. His education and experience include:

2018 – 2021 – Ph.D., Heidelberg University (Statistical Modeling in Psychology)
2021 – 2023 – Postdoctoral Researcher, Heidelberg University (Deep Learning)
2023 – present – Assistant Professor, Rensselaer Polytechnic Institute

Title: Toward Foundation Agents: How Multimodal Models Learn (and Fail to Learn) the Physical World

Abstract: Today’s multimodal models are often trained with a brute-force “align everything” recipe, yet it is still unclear how cross-modal intelligence can emerge. We argue the key question is mechanistic: how can models go beyond static alignment annotations to learn from physical-world interaction and support goal-directed decision making? We systematically study multimodal learning through the MDP agent loop: state estimation, world modeling for planning, and control for safety. First, we open up the black box, and intervene inside embeddings to reveal how geometry is lost, and design ways to retain geometric structure. Second, we inject world-model priors to teach dynamics through RAGEN/VAGEN, enabling multi-step planning rather than token matching. Third, we introduce ODE-Steer for safe agents, which steers internal activations into “safe zones” where reasoning stays reliable and controllable. Lastly, we lay out the future that true multimodal intelligence requires more than aligning tokens; it requires aligning the internal mechanisms of the model with the geometry of the world.

Bio: Manling Li is an Assistant Professor at Northwestern University and an Amazon Scholar. She was a postdoc at Stanford University, and obtained Ph.D. degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on Reasoning, Planning and Compositionality, in the intersection of Language, Vision, and Robotics. Her work has been recognized as ACL 2025 Inaugural Dissertation Award Honorable Mention, MIT Tech Review Innovators Under 35, ACL’24 Outstanding Paper Award, NAACL'21 Best Demo Paper Award, ACL'20 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc. She served as virtual chairs of ACL 25, publication chairs at NAACL 25, demo chairs at EMNLP 24, etc. Additional information is available at https://limanling.github.io/.

Date: Wednesday, March 4, 2026, at 12:00 p.m.

Location: Coda, Room 230 (Google Maps link)

Lunch provided!

Title: From Sparse Sensors to Continuous Fields: STRIDE for Spatiotemporal Reconstruction

Abstract: We introduce STRIDE (Spatio-Temporal Recurrent Implicit DEcoder), a novel deep learning framework for reconstructing high-dimensional spatiotemporal fields from sparse point-sensor measurements. Existing approaches often struggle to generalize across trajectories and parameter settings, or rely on discretization-tied decoders that do not naturally transfer across meshes and resolutions. Our proposed approach is a two-stage framework which maps a short window of sensor measurements to a latent state with a temporal encoder and reconstructs the field at arbitrary query locations with a modulated implicit neural representation (INR) decoder. Using the Fourier Multi-Component and Multi-Layer Neural Network (FMMNN) as the INR backbone improves representation of complex spatial fields and yields more stable optimization than sine-based INRs. We provide a conditional theoretical justification: under stable delay observability of point measurements on a low-dimensional parametric invariant set, the reconstruction operator factors through a finite-dimensional embedding, making STRIDE-type architectures natural approximators. Experiments on four challenging benchmarks spanning chaotic dynamics and wave propagation show that STRIDE outperforms strong baselines under extremely sparse sensing, supports super-resolution, and remains robust to noise.

Bio: Yanjie Tong is a second-year CSE Ph.D. student advised by Dr. Peng Chen. He earned both his B.S. in Mathematics and Physics and his B.Eng. in Energy and Power Engineering at Tsinghua University. His research focuses on methods for learning latent dynamics in low-dimensional spaces and their application to real-world problems.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

The world around us is inherently 3D, and this structure is essential from engineering and the physical sciences to photogrammetry and robotics. Yet computing and learning with real-world geometric data remains remarkably difficult, stubbornly stuck on challenges of robustness, representation, and scale. Noisy scans, mismatched representations, and massive datasets make even basic operations fragile and slow, while learning with 3D data demands a new set of architectures purpose-built for the task.

This talk will present methods that unlock 3D data for reliable processing and learning in the wild. I'll cover neural network architectures designed specifically for 3D data to offer physical consistency and geometric fidelity by-construction, simulation techniques that operate natively on scanned environments, and the foundational geometric algorithms that underpin it all. Throughout, a key theme is interfacing principled geometric methods with modern machine learning.

Bio

Nicholas Sharp's research develops algorithms and architectures to make computing with geometric data easy, efficient, and reliable. He received his PhD in Computer Science from Carnegie Mellon University advised by Keenan Crane, and completed his postdoc at the University of Toronto & Fields Institute for Mathematics with Alec Jacobson. Currently, he is a Senior Research Scientist at NVIDIA, leading research on 3D learning that powers applications from autonomous driving to generative AI. His work has won multiple best paper awards (SIGGRAPH '22, SIGGRAPH Asia '23, SGP '20 & '25), developed widely-used software including the award-winning Polyscope library, and has been covered in the press by the New York Times and Ars Technica. For more, see www.nmwsharp.com.

Bio

Yifei Li is a Ph.D. candidate in EECS at MIT, advised by Prof. Wojciech Matusik. Her research bridges high-fidelity physics simulation with machine learning to enable systems that reason through physical environments, powering applications across computational design, digital twins, assistive robotics, and biomedical engineering. Yifei received her B.S. in Computer Science from Carnegie Mellon University. Her professional experience includes research internships at Meta Reality Labs, NVIDIA, the Boston Dynamics AI Institute, and Facebook AI Research. She is the recipient of the MIT Stata Family Presidential Fellowship and has been named a Rising Star in both Computer Graphics and EECS.

Physics simulation increasingly sits in the loop of interactive content creation, digital twins, and embodied AI. A central dilemma cuts across these applications: the tradeoff between simulation speed and fidelity, especially at scale. In practice, an equally limiting barrier is interactive control. Artists and designers need coarse previews they can manipulate in real time, with confidence that refinement will not overturn their decisions. Standard multiresolution approaches often fall short because coarse solutions can drift, and refinement can introduce surprises due to numerical stiffening, geometric inconsistency, and contact.

To address this challenge, my research program introduces Progressive Simulation, a general-purpose, coarse-to-fine level-of-detail (LOD) framework for physics simulation that brings and generalizes ideas from progressive rendering and progressive geometry to the simulation setting. The core idea is to produce a fast, coarse, artifact-free, predictive preview under tight computational budgets for rapid iteration, then apply consistent refinement across resolutions that converges to a richly detailed, high-fidelity result as more resources or stricter accuracy requirements become available. In this talk, I will present methods that realize this approach in both progressive quasistatics and progressive dynamics, with results spanning cloth, shells, and volumetric objects, enabling experimental design, interactive staging, and physics-based storytelling. Finally, I will outline future directions toward broader applicability, including multi-fidelity, multi-physics simulation systems powered by scalable solvers.

Bio

Jiayi (Eris) Zhang is a fifth-year PhD candidate in Computer Science at Stanford University, advised by Prof. Doug James, and works closely with Dr. Danny Kaufman at Adobe. She received her BS in Computer Science and Mathematics from the University of Toronto, where she conducted research with Prof. Alec Jacobson. Her research spans physics simulation, geometry processing, numerical optimization, and spatial intelligence, with a focus on building algorithms, models, and tools that enhance human productivity and creativity in science, engineering, and interactive authoring workflows. She is a Stanford Reed-Hodgson Fellow, a WiGRAPH Rising Star, and an MIT EECS Rising Star, and has received the Adobe Women-in-Technology Scholarship, the Roblox Graduate Fellowship, and the NVIDIA Graduate Fellowship. She has also interned at Adobe and NVIDIA across multiple summers.

Title: Symbolic Abstractions Enable Exascale Simulations of Reacting Flows and Automatic Differentiation

Abstract: Accurate simulations of reacting flows require a computational representation of the chemical source terms. Dedicated libraries can provide such representations, acting as interfaces between mechanistic data and flow solvers. However, libraries form a fragmented computational landscape, designed to satisfy one application (e.g., combustion or plasma chemistry) and target specific computing hardware. These considerations preclude the simulation of multi-chemistry flows such as plasma-coupled combustion, automatic differentiation, and joint optimization of flow solvers on modern hardware such as GPUs. A symbolic representation of the chemistry is developed to address these challenges. At the symbolic level, conflicts across applications and hardware vanish. The representation is language and library agnostic, transformed through code generation to enable performant GPU execution and automatic differentiation. The Pyrometheus suite implements the proposed approach. Its capabilities and performance are demonstrated for combustion, plasma, and thermonuclear chemistry. The implementation supports existing HPC frameworks, enabling exascale simulations of reacting flows with the state-of-the-art flow solver MFC.

Bio: Esteban Cisneros–Garibay is an assistant professor at the University of Tennessee Space Institute. His research focus is on high-speed chemically-reacting turbulent flows, using computational methods. He obtained his PhD in Theoretical & Applied Mechanics at The University of Illinois Urbana–Champaign in 2021. From 2021 to 2022, he was Presidential Postdoctoral Fellow at Princeton University, and then a data scientist at Bayer AG until 2025. 

World models that recreate and simulate the physical world hold transformative potential across robotics, entertainment, and engineering analysis. Achieving this vision requires both generating 3D environments from limited observations and predicting how they evolve under physical actions. Pure physical modeling provides guarantees and action control but demands complete state specification rarely available in practice; pure generative learning handles incomplete information to produce realistic content but lacks the structured representations needed for physical interaction and reasoning. This talk presents physics-grounded world models, integrating these approaches to leverage their complementary strengths: physical representations provide the structured interface for actions and consistency guarantees, while generative models supply visual realism and compensate for incomplete observations. I will demonstrate this framework across two core capabilities---generating 3D worlds from single images and simulating dynamics under physical actions---and show how it extends to real engineering problems in fluid and thermal analysis.

Bio

Hong-Xing (Koven) Yu (https://kovenyu.com/) is a PhD candidate at the Computer Science Department of Stanford University, advised by Prof. Jiajun Wu. His research focuses on physics-grounded world models. He is a recipient of the SIGGRAPH Asia Best Paper Award, the Stanford SoE Fellowship, the Qualcomm Fellowship, and the Meshy Fellowship, and a finalist of the NVIDIA Fellowship, the Meta Fellowship, the Jane Street Fellowship, and the Roblox Fellowship.

Speaker:  Shanto Rahman, Ph.D. Candidate, University of Texas at Austin 

Abstract: 

Software testing is one of the most scalable ways to check program correctness. However, nondeterminism in test execution and rapid code evolution can cause tests to break and produce false failures, where the same test may fail without any underlying bug. As these false failures accumulate, developers may start ignoring test results, allowing real bugs to reach production. My research aims to make software testing reliable by detecting, understanding, reproducing, and repairing such broken tests. In this talk, I first present my work on predicting and diagnosing the root causes of flaky tests using context-aware attribution, helping developers understand why nondeterministic failures occur while reducing runtime and memory cost. Next, I introduce automated repair techniques for both nondeterminism- and evolution-induced test breakage using intent-preserving dynamic instrumentation and generative AI. These techniques are evaluated on large-scale, real-world datasets, and achieve high repair success. I conclude by outlining a path from reliable test diagnosis and repair, extending these reliability foundations to modern computing systems for trustworthy, safe operation in practice. 

Bio: 

Shanto Rahman is a final-year Ph.D. candidate at The University of Texas at Austin, advised by August Shi. Her research sits at the intersection of software engineering, program analysis, and AI, with a focus on making software testing reliable under nondeterminism and continuous code evolution. Her work has been published in top venues including ICSE, OOPSLA, ASE, and ICST. Shanto has gained industry experience through research internships at Google and Amazon Web Services (AWS). She has been recognized as an MIT EECS Rising Star and has received multiple UT Austin fellowships and awards. 

As we move toward the era of physical AI systems that perceive, reason, and act, the demand for virtual worlds that behave like reality has never been higher. My work targets this challenge through 'Physically Grounded Visual Computing': a perspective that bridges high-fidelity computer graphics engines with modern 3D vision, generative AI, and robots. I will present our breakthroughs in robust graphical simulations, specifically showing how continuum mechanics and contact mechanics provide the mathematical guarantees needed to model complex phenomena, from glacial calving to tactile manipulation. Beyond dynamic simulations, I will discuss how we are fusing these engines with generative models, such as physics-informed 3D Gaussian Splatting, video diffusion, and 3D reconstruction models, to create digital twins that look photorealistic and act physically valid. We build towards a roadmap for using these physics-based world engines to generate synthetic data at scale and equip agents with the physical intuition they need to succeed in the real world.

Bio

Chenfanfu Jiang (PhD’15) is a Professor of Mathematics and the director of the Artificial Intelligence and Visual Computing (AIVC) Lab at UCLA. He primarily works on computer graphics and its interdisciplinary synergies with 3D vision, generative AI and physical AI. He has led the development of widely-adopted simulators with Particle-In-Cell, Material Point, and Incremental Potential Contact methods. He received the dissertation award from UCLA, NSF CAREER, and multiple faculty awards from Amazon, Style3D, and Sony, as well as best paper awards/honorable mentions at SIGGRAPH, SCA, MIG, ICRA, and IROS.

Speaker:  Sourav Chakraborty, Professor, Indian Statistical Institute

Abstract: 

The problem of testing whether an input satisfies a given property can be significantly easier than estimating the input’s distance from having that property. However, for certain classes of properties, these tasks may be quite similar.

One of the newer models for testing distribution properties is the huge object model, introduced by Goldreich and Ron in 2023. In this talk, we investigate the query complexity of testing index-invariant properties within this model. We adapt Szemerédi’s regularity method to this setting and prove that, for index-invariant properties, constant query testability implies constant-query estimability.

This talk is based on the following works:

1. “Testing vs Estimation for Index-Invariant Properties in the Huge Object

Model,” joint work with Eldar Fischer, Arijit Ghosh, Amit Levi, Gopinath Mishra,

and Sayantan Sen, STOC 2025.

2. “Testing of Index-Invariant Properties in the Huge Object Model,” joint work

with Eldar Fischer, Arijit Ghosh, Gopinath Mishra, and Sayantan Sen, COLT 2023

Bio:

Sourav Chakraborty is a Professor in the Advanced Computing and Microelectronics Unit (ACMU) of the Computer and Communication Sciences Division (CCSD) at the Indian Statistical Institute (ISI), Kolkata, India. Before joining ISI in July 2018, he was a faculty member at the Chennai Mathematical Institute, India, from September 2010. He completed his PhD in Computer Science in June 2008 at the University of Chicago under the supervision of Prof. László Babai and did postdoctoral stints at Technion, Israel, and CWI, Amsterdam.

Speaker:  Courtney Miller, Ph.D. Candidate, Carnegie Mellon University 

Abstract: 

Software development is shaped by constant disruption—from open-source dependency abandonment to generative AI (GenAI) tools fundamentally reshaping how code is written. 
 
In this talk, I'll present two lines of work, each examining a different disruption. First, I'll share my work on the disruption of open source dependency abandonment, revealing the unsupported challenges developers face when identifying and responding to abandonment, identifying factors that support successful downstream response, and demonstrating how theory-driven LLM-based systems can make those risks more visible. Second, I’ll present findings from my recent collaboration with Microsoft Research about the disruption of GenAI adoption on development teams, demonstrating why adoption succeeds for some developers but stalls for others in seemingly similar contexts, showing the amplification effect organizational support mechanisms can have, and introducing the Productivity Pressure Paradox: a dynamic where increased productivity expectations without corresponding support can actually prevent developers from building the skills that would deliver those gains. 
 
Finally, I’ll outline my research vision for helping developers anticipate and navigate the disruptions of tomorrow—from building AI-powered tools for software supply chain resilience to studying how GenAI is reshaping development team practices and processes. 

Bio:

Courtney Miller is a Software Engineering Ph.D. Candidate in the School of Computer Science at Carnegie Mellon University, co-advised by Bogdan Vasilescu and Christian Kästner. Her research examines how developers navigate technological disruptions—from dependency abandonment threatening open source software supply chain sustainability and security to generative AI adoption reshaping development workflows. 
 
Her empirical approach combines human-centered qualitative techniques with large-scale data-driven analysis, modeling, and visualization to inform the development of theoretically-grounded insights and practice-driven tooling solutions. As part of the Secure Software Supply Chain Center (S3C2), her work has informed security practices through contributions to federal agency and industry supply chain security summits. 
 
Courtney has earned three ACM SIGSOFT Distinguished Paper Awards at the International Conference on Software Engineering (ICSE) and an NSF Graduate Research Fellowship. She has conducted two research internships at Microsoft Research. She holds a B.A. with honors in Computer Science and Statistics from New College of Florida. In her free time, she enjoys indoor cycling and leisurely walks with her 14-year-old toy poodle Chanel. 

Date: Tuesday, February 17, 2026 at 11:00 am

Location: Coda Building, Room 114 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Coffee and snacks provided!

Title: Uncompromising Performance with Exocompilation

Abstract: Performance is the currency of modern computing. Achieving peak throughput on fast‑evolving accelerators demands more control than today's compilers provide. I will present Exo and the Exocompilation paradigm: a user‑schedulable programming language that shifts two responsibilities traditionally hard‑coded in compilers--hardware backends and optimization strategies--into safe, extensible user libraries. In Exo, optimizations are expressed as verified rewrites that guarantee functional equivalence, and low‑level primitives (e.g., explicit instruction selection, GPU asynchrony) give programmers precise control. This design enables targeting CPUs, GPUs, and matrix engines, and supports concise, reusable scheduling libraries that match or surpass the performance of state-of-the-art libraries such as cuBLAS and MKL. Exo‑generated kernels already ship at scale, including on Apple devices.

Bio: Yuka Ikarashi is a sixth-year Ph.D. candidate at MIT CSAIL. She received her M.S. from MIT in 2022 and her B.S. from the University of Tokyo in 2020. Passionate about compilers and programming languages for high-performance computing, she created the Exo programming language. She has previously worked at Apple, Amazon, and CERN, applying her research to various accelerators and applications. She has been awarded the Quad Fellowship, the Masason Foundation Fellowship, the Funai Foundation Fellowship, the ML and Systems Rising Stars Award, and the Rising Stars in EECS Award. Originally from Tokyo, Japan, she is an avid traveler and has visited 36 countries.

Title: Expediting Next-Generation AI for Health via KG and LLM Co-Learning

Abstract: Large language models (LLM) have brought disruptive progress to information technology from accessing data to performing analytical tasks. While demonstrating unprecedented capabilities, LLMs have been found unreliable in tasks requiring factual knowledge and rigorous reasoning, posing critical challenges in domains such as healthcare. Knowledge graphs (KG) have been widely used for explicitly organizing and indexing biomedical knowledge, but the quality and coverage of KG are hard to scale up given the notoriously complex and noisy healthcare data with multiple modalities from multiple institutions. Existing approaches show promises in combining LLMs and KGs to enhance each other, but they do not study the techniques in real healthcare contexts and scenarios. In this talk, I will introduce our research vision and agenda towards KG-LLM co-learning for healthcare, followed by success examples from our recent exploration on LLM-aided KG construction, KG-guided LLM enhancement, and federated multi-agent systems. I will conclude the talk with discussions on future directions that can benefit from further collaborations with researchers interested in data mining or biomedical informatics in general.

Bio: Carl Yang is an Assistant Professor of Computer Science at Emory University, jointly appointed in the Rollins School of Public Health and Nell Hodgson Woodruff School of Nursing. He received his Ph.D. in Computer Science at University of Illinois, Urbana-Champaign in 2020, and B.Eng. in Computer Science and Engineering at Zhejiang University in 2014. His research areas span data mining, deep learning, multimodality foundation models and trustworthy AI, with applications in graph analytics, neuroscience, biomedicine and healthcare. Carl's research results have led to 250+ peer-reviewed publications in top venues across AI/ML and medicine/healthcare. He serves as the Organizer of Health Day and Inaugural Program Chair of AI4Sciences in KDD, and the Chair-Elect of AMIA KDDM Working Group. He is also a 1% NIH/CDC Grant Awardee in 2025, and a recipient of the ACM SIGKDD Rising Star Award in 2025 (one winner a year in the world), NSF CAREER Award in 2025, NIH K25 (Career) Award in 2023, and multiple Best Paper Awards such as of MedInfo 2025, KDD Health Day 2022, ML4H 2022, and ICDM 2020.

Date: Thursday, February 12, 2026 at 11:00 am

Location: Coda Building, Room 114 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Coffee and snacks provided!

Title: Discovering Model Manifolds of Emergent Function from High-Dimensional Data

Abstract: Natural systems of many interacting constituents often organize to produce low-dimensional emergent behaviors. For example, hundreds of thousands of neurons in the brain coordinate to represent and navigate the three-dimensional world around us. Similar emergent behaviors arise in the tens of thousands of genes that coordinate to produce the cellular processes that maintain life. Recent technological advances now allow us to simultaneously measure the activity of tens of thousands of neurons in the brain, and the expression of every gene at the resolution of single cells, giving us the opportunity to model emergent function at an unprecedented scale. How do we take such high dimensional data and discover interpretable and quantitative models of the algorithms that brains use to navigate the world, and cells use to achieve life? 

In this talk, I will discuss novel methods I am pioneering that combine differential geometry with deep neural networks to quantitatively model emergent behaviors using high-dimensional, high-throughput data from neuroscience and genomics. I will discuss fundamental problems of model identifiability (degeneracy in model parameters that fit the data), interpretability (mapping discovered emergent variables back to neuron and gene circuits), and predictive power, and how geometry-aware deep learning models ameliorate these problems to enable new discoveries. Finally, I will describe some of the discoveries that we have made using this method about how the neural code is structured and used to accurately represent and efficiently search mental models of the world in mice learning to navigate a maze.

Bio: I am interested in discovering and modeling how articulate and intelligent biological function emerges from simpler parts. My undergraduate degree was in Biomedical Engineering at Duke University, where I did single-unit recordings in behaving macaques under electromagnetic stimulation. I did my graduate work in Bioengineering at the University of Pennsylvania designing articulate and intelligent biological function in mechanistic mathematical models ranging from cooperative energy landscapes in mechanical models of proteins, to mental simulation of video games in dynamical models of the brain. I am currently a postdoctoral fellow in Theoretical Physics, Neurotech, and AI for Science at Cornell University developing novel techniques to discover quantitative, interpretable, and predictive models of the algorithms that brains use to navigate physical and conceptual spaces, and cells use to produce life.

Date: Thursday, February 19, 2026 at 11:00 am

Location: Coda Building, Room 230 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Coffee and snacks provided!

Title: Towards a Less Conservative Theory of Machine Learning: Unstable Optimization and Implicit Regularization

Abstract: Deep learning’s empirical success challenges the “conservative" nature of classical optimization and statistical learning theories. Classical theory mandates small stepsizes for training stability and explicit regularization for complexity control. Yet, deep learning leverages mechanisms that thrive beyond these traditional boundaries. In this talk, I present a research program dedicated to building a less conservative theoretical foundation by demystifying two such mechanisms: 

1. Unstable Optimization: I show that large stepsizes, despite causing local oscillations, accelerate the global convergence of gradient descent (GD) in overparameterized logistic regression. 

2. Implicit Regularization: I show that the implicit regularization of early-stopped GD statistically dominates explicit $\ell_2$-regularization across all linear regression problem instances.

I further showcase how the theoretical principles lead to practice-relevant algorithmic designs (such as Seesaw for reducing serial steps in large language model pretraining). I conclude by outlining a path towards a rigorous understanding of modern learning paradigms.

Bio: Jingfeng Wu is a postdoctoral fellow at the Simons Institute for the Theory of Computing at UC Berkeley. His research focuses on deep learning theory, optimization, and statistical learning. He earned his Ph.D. in Computer Science from Johns Hopkins University in 2023. Prior to that, he received a B.S. in Mathematics (2016) and an M.S. in Applied Mathematics (2019), both from Peking University.

Date: Thursday, February 5, 2026 at 11:00 am

Location: Coda Building, Room 114 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Coffee and snacks provided!

Title: Designing Dynamic Measure Transport for Density-Driven Sampling

Abstract: Sampling from a target probability distribution is fundamental to modern computational science and machine learning. Sampling is the essence of Monte Carlo integration, enables uncertainty quantification in Bayesian inference, and underlies generative models that have the ability to synthesize convincing text, images, and far beyond. A powerful, emerging approach to sampling is dynamic measure transport (DMT): the idea is to design an ordinary or stochastic differential equation that evolves samples from a tractable reference distribution (e.g., a Gaussian) to the desired target distribution. DMT is state-of-the-art in generative modeling and underlies techniques such as diffusion models and flow-matching, but DMT pipelines for density-driven sampling tasks, as arising in computational chemistry and Bayesian inference, are significantly less developed. In this talk, I will discuss my work to make density-driven DMT a reality via: (1) development of new, gradient-free particle systems for Bayesian sampling, (2) principled design of DMT via PDE-constrained optimization, and (3) scalability through the exploitation of sparse conditional dependence structure. I will describe how these efforts will enable new DMT approaches to complex sampling problems--such as ensemble data assimilation, in which the prior is only known through samples---and sketch future work on stochastic inverse problems, in which an unknown distribution must be recovered from indirect measurements.

Bio: Aimee Maurais is in the final year of her Computational Science and Engineering Ph.D. at MIT. Prior to beginning her graduate work, Aimee earned bachelor’s degrees in Mathematics and Computational Modeling and Data Analytics from Virginia Tech and spent 1.5 years on the technical staff at MIT Lincoln Laboratory. In her research, Aimee makes use of geometry and structure to develop principled computational methods for probabilistic modeling and inference, and her current focus is on dynamic measure transport for sampling. Aimee’s work is supported by fellowships from the National Science Foundation and Google.

 1 p.m. "Modernizing C4G BLIS: The Road to BLIS 4.0"

Mitchell Rysavy (Senior Software Engineer, Microsoft)

 1:30 p.m."Computing for Good in an Increasingly Bad World"

Michael Best (Professor, School of International Affairs and School of Interactive Computing, Georgia Tech)

 2 p.m. "Disaster Response in Rural Communities"

Justin McLellan (Software Engineering Manager, Crexi)

 2:30 p.m. "HandS together: Innovation and Collaboration for Greater Impact"

Protip Biswas (Vice President, United Way Regional Commission on Homelessness)

Title: Learning World Models and Agents for High-Cost Environments

Abstract: While neural networks have achieved superhuman performance in domains with low-cost simulations—from AlphaGo to LLMs—their application to the physical world is bottlenecked by a fundamental challenge: high-cost interactions. In fields like robotics, ML engineering, and the natural sciences, every action or experiment is expensive and time-consuming. This talk outlines strategies for building intelligent agents that learn efficiently despite these real-world constraints. We first address the physical world by showing how learned world models can serve as high-fidelity simulators for robotics, enabling extensive policy refinement before deployment on costly hardware. We then turn to complex engineering domains, where actions like running an ML program incur significant time delays, and discuss adaptations to reinforcement learning to make it robust for these long action settings. Finally, we show how compositional generative models can navigate the vast hypothesis spaces in science, intelligently proposing experiments to accelerate the pace of discovery.

Bio: Sherry Yang is an Assistant Professor of Computer Science at NYU Courant and a Staff Research Scientist at Google DeepMind. She researches in machine learning with a focus on reinforcement learning and generative modeling. Her current research interests include learning world models and agents, and their applications in robotics and AI for science. Her research has been recognized by the Best Paper award at ICLR and various media outlets such as VentureBeat and TWIML. She has organized tutorials, workshops, and served as Area Chairs at major conferences (NeurIPS, ICLR, ICML, CVPR). Prior to her current role, she was a post-doc at Stanford working with Percy Liang. She received her Ph.D. from UC Berkeley advised by Pieter Abbeel and master’s and bachelor's degrees from MIT.

Program Courses:

• CS 4001 – Computing & Society (Ethics Credit) - Students will critically examine how issues around technology, such as globalization, bias, privacy, and others, affect people worldwide.

• CS 3751 – Intro to User Interface Design - Students will explore culturally situated approaches to design and, through the course project, create an inventive design artifact.

Applications are now open HERE. Email indiaprogram@cc.gatech.edu for questions and to find out more about upcoming Info Sessions.

Program Courses:

• CS 4001 – Computing & Society (Ethics Credit) - Students will critically examine how issues around technology such as globalization, bias, privacy, and others affect people worldwide.

• CS 3751 – Intro to User Interface Design - Students will explore culturally situated approaches to design and, through the course project, create an inventive design artifact.

Applications are now open HERE. Email indiaprogram@cc.gatech.edu for questions and to find out more about upcoming Info Sessions.

Title: Accelerating Fundamental Physics Research with AI: From Designing Theories to Testing Them in Experiments

Abstract: In fundamental physics, we build bottom-up theories to explain the universe from elementary particles to the nature of gravity. Theoretical advancements have traditionally relied on human insight into complex mathematics without a systematic way to explore ideas. I will talk about an emerging research direction in neuro-symbolic AI to systematize this exploration and discuss challenges in scaling this approach.

Once you have a set of interesting theories, testing them in experiments, such as at the Large Hadron Collider, is yet another computational and engineering challenge. High-dimensional data are generated at 50 MHz (40 TB/s), and we need to make decisions using efficient hardware. I will also discuss how neural inference techniques let us maximally use the high-dimensional data and current efforts to make these algorithms compute-efficient.

Bio: Aishik Ghosh is an assistant professor in the School of Physics at Georgia Tech with a focus on developing AI methods to accelerate fundamental physics and astrophysics. His group works on theoretical physics, statistical methods, and experiment design. For robust scientific applications, Dr. Ghosh focuses on uncertainty quantification, interpretability, and verifiability of AI algorithms, targeting publications in physics journals and ML conferences.

Program Courses:

• CS 4001 – Computing & Society (Ethics Credit) - Students will critically examine how issues around technology such as globalization, bias, privacy, and others affect people worldwide.

• CS 3751 – Intro to User Interface Design - Students will explore culturally situated approaches to design and, through the course project, create an inventive design artifact.

Applications are now open HERE. Email indiaprogram@cc.gatech.edu for questions and to find out more about upcoming Info Sessions.

Title: Generative Artificial Intelligence for Uncertainty Quantification

Abstract: Generative artificial intelligence models, including variational autoencoders, normalizing flows, generative adversarial networks, and diffusion models, have dramatically advanced the realism and quality of generated images, text, and audio. Beyond media synthesis, these models hold great promise as powerful tools for probability density estimation and high-dimensional sampling, which are central to uncertainty quantification (UQ) tasks such as amortized Bayesian inference and data assimilation. However, while research on image and audio synthesis emphasizes producing high-quality individual samples, UQ applications demand accurate approximation of statistical quantities of interest rather than visually realistic samples. As a result, direct application of existing generative models to UQ problems can lead to biased approximations or unstable training. In this talk, I will introduce several new generative approaches tailored to UQ. These include training-free diffusion models for density estimation, a score-based nonlinear filter for data assimilation, and conditional diffusion models for amortized Bayesian inference. I will demonstrate their effectiveness across a range of tasks, including density estimation for unimodal and multimodal distributions, learning stochastic dynamical systems, parameter estimation via amortized inference, and scalable data assimilation for atmospheric models.

Bio: Dr. Guannan Zhang is a Distinguished Staff Scientist in Computer Science and Mathematics Division at Oak Ridge National Laboratory (ORNL). He earned my Ph.D. in applied mathematics at Florida State University in 2012. He joined ORNL in 2012 as the Householder fellow. He received the DOE Early Career Award in 2022. He has been holding a joint faculty appointment with the Department of Mathematics and Statistics at Auburn University since 2014, and a joint faculty appointment with Department of Mathematics at University of Tennessee since 2022. Guannan's research interests include high-dimensional approximation, uncertainty quantification, machine learning and artificial intelligence, stochastic methods for scientific inverse problems. 

We invite you to study abroad this summer with the College of Computing! Take 12 GT credit hours while immersing yourself in the exciting environments of Barcelona. See below for information on program details and application info.

2026 Barcelona Summer Program course offerings include:

• CS 2050 – Intro to Discrete Math for CS
• CS 3750 – Human-Computer Interface Design & Evaluation
• CS 4001 – Computing and Society (Ethics credit)
• CS 4400 – Intro to Database Systems
• CS 4660 – Intro to Educational Technology
• CS 4726 – Privacy, Technology, Policy, & Law (Ethics Credit)
• CS 4803-EDG – Educational Game Design
• ARCH 3135 – City Literacy: What Makes Great Cities Great (Social Sciences Credit)

Review the attached flyer and email barcelona@cc.gatech.edu for questions. Application deadline: December 1st in Atlas.

We invite you to study abroad this summer with the College of Computing! Take 12 GT credit hours while immersing yourself in the exciting environments of Berlin. See below for information on program details and application info.

2026 Berlin Summer Program course offerings include:

• CS 2340 – Objects & Design
• CS 2701 – Startup Lab
• CS 3510 – Algorithm Design
• CS 3600 - Intro to Artificial Intelligence
• CS 4001 – Computing and Society (Ethics credit)
• CS 4400 – Intro to Database Systems
• LMC 3262 – Performance Studies (Humanities credit)
• LMC 3314 – Technologies of Representation (Humanities credit)

Berlin is a growing Tech Hub in Europe, home to innovative companies and research institutions, and host to tech meetups, hackathons, and conferences. With focus on entrepreneurship and start-up culture, this program is a great option if you are a CM major, if your CS Threads are Intelligence or Information-Internetworks, or if you’re doing the Create-X: Idea-To-Protype Junior Design option.

Review the attached flyer and email berlin@cc.gatech.edu for questions. Application deadline: December 1st in Atlas.

Date and Time: Nov. 13 at 11 a.m.

Location: KACB 2443

Title: Distributed Optimization and Consensus in the Presence of Adversarial Agents 

Abstract: Due to their many potential applications, there is a rich history of research on distributed consensus and distributed optimization. Distributed consensus requires the agents in the system to reach agreement on an output as a function of the inputs of the various agents. Distributed optimization requires the agents to collaboratively optimize a global cost function that consists of the summation of the local cost functions of the individual agents. Loss minimization in machine learning is a special case of this problem, and it has many other applications as well. The problems of consensus and optimization are inter-related, with some commonalities in the solutions used for these problems.

In this talk, we will consider the above two problems in the presence of adversarial agents. The adversarial agents may misbehave and attempt to tamper with the outcome of the distributed computation. Such adversarial behavior may occur due to security compromise of the agents or due to software/hardware failure. We will take an overview of some recent results on the above two problems, with an intuitive presentation of the proposed solutions.

The talk will NOT assume any distributed computing background from the audience.

Bio: Nitin Vaidya is the Robert L. McDevitt, K.S.G., K.C.H.S. and Catherine H. McDevitt L.C.H.S. Chair Professor Computer Science at Georgetown University, where he served as the Department Chair during 2018-24. His current research interests are in the area of distributed algorithms, and previously he has worked on wireless networks. He received a Ph.D. from the University of Massachusetts at Amherst. He previously served as a Professor and Associate Head in Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. He has co-authored papers that received awards at several conferences, including SSS, ACM MobiHoc and ACM MobiCom. He is a fellow of the IEEE. He has served as the Chair of the Steering Committee for the ACM PODC conference, as the Editor-in-Chief for the IEEE Transactions on Mobile Computing, and as the Editor-in-Chief for ACM SIGMOBILE publication MC2R.

In this talk, Dr. Craig Zilles will:

• Share research on frequent and “second-chance” testing to enhance learning and reduce anxiety
• Discuss patterns of cheating on unproctored online assessments
• Describe how the CBTF reduces faculty workload while maintaining exam integrity

Dr. Zilles will also provide practical insights for implementing computer-based testing facilities at other institutions and discuss the potential benefits for large STEM courses.

Speaker Bio:
Craig Zilles is a Professor and Severns Faculty Scholar in the Siebel School of Computer and Data Science at the University of Illinois at Urbana-Champaign. His current research focuses on applying computing and data analytics to education, including the development of the Computer-Based Testing Facility (CBTF).  Previously, his research focused on the interaction between compilers and computer architecture, and he developed the first algorithm that allowed rendering arbitrary three-dimensional polygonal shapes for haptic interfaces (force-feedback human-computer interfaces).  He received the IEEE Education Society's 2010 Mac Van Valkenburg Early Career Teaching Award and an NSF CAREER award.  At Illinois, he has received a wide range of teaching awards, including a 2018 Campus Award for Excellence in Undergraduate Teaching, a 2013 Illinois Student Senate Teaching Excellence Award, and the College of Engineering's Rose Award (2007) and Everitt Award (2008) for Teaching Excellence.  He holds 5 patents and his research has been recognized by a best paper awards from ASPLOS in 2010 and 2013 and by selection for inclusion in the IEEE Micro Top Picks from the 2008 Computer Architecture Conferences.

Date & Time:
Tuesday, November 11, 2025 — 10:00 a.m

Hosted by:
School of Computing Instruction, College of Computing

Title: Understanding the Genetic Basis of Complex Traits and Common Diseases from Biobank-Scale Data: Statistical and Computational Challenges

Abstract: The quest to understand the interplay between evolution, genes and traits has been revolutionized by the collection of rich phenotypic and genetic data across millions of individuals in diverse populations. However, analyses of these Biobank-scale datasets present substantial statistical and computational challenges. 

I will describe how we bring together statistical and computational insights to design accurate and highly scalable algorithms for a suite of problems that arise in the analysis of Biobank data: highly scalable randomized inference algorithms to dissect the genetic architecture of complex traits and deep-learning based phenotype imputation to deal with complex patterns of missingness. By applying these methods to about half a million individuals from the UK Biobank, we obtain novel insights how genetic effects are distributed across the genome, the relative contributions of additive, dominance and gene-environment interaction effects to trait variation, and new genes that confer risk for hard-to-measure diseases.

Bio: Sankararaman is a professor in the Departments of Computer Science, Human Genetics, and Computational Medicine at UCLA. His research interests lie at the interface of computer science, statistics, and biology. His lab develops machine learning algorithms to analyze genomic and biomedical data with the broad goal of understanding the interplay between evolution, genomes, and traits. His work has led to methods for identifying disease genes in diverse human populations, discovered  interbreeding between archaic and modern humans, established guidelines for sharing genetic data while maintaining privacy, and characterized the genetic architecture of human complex traits.

He received a B.Tech. in Computer Science from the Indian Institute of Technology, Madras, a Ph.D. in Computer Science from UC Berkeley and was a postdoctoral fellow in Harvard Medical School before joining UCLA. He is a recipient of a NSF Career Award, a NIH Pathway to Independence Award, fellowships from Microsoft Research, the Sloan Foundation, the Okawa Foundation and the Simons Institute , and multiple teaching awards.

Title: Interpretable AI: A Goal-Oriented, User-Centric Perspective

Abstract: How do deep learning “vision” models arrive at their predictions? This fundamental question has motivated a vibrant line of work in explainable and interpretable AI (XAI). In this talk, I will introduce XAI from a goal-oriented, user-centric perspective, emphasizing explanations that serve scientists’ and practitioners’ needs rather than merely optimizing proxy metrics. I will then showcase milestones from Imageomics (https://imageomics.osu.edu/) -- an emerging interdisciplinary field to make biological traits computable from images—including Finer-CAM and interpretable transformers for precise, fine-grained attribution; sparse autoencoders (SAE) that uncover meaningful features for scientific discovery; BioCLIP, a vision foundation model built for the Tree of Life; and Static Segmentation by Tracking (SST) for specimen image analysis. I will conclude with a discussion on the future research opportunities in this exciting frontier.

Bio: Wei-Lun (Harry) Chao (https://cse.osu.edu/people/chao.209) is an Associate Professor and College of Engineering Innovation Scholar in Computer Science and Engineering at The Ohio State University (OSU). His research focuses on machine learning and computer vision, with applications spanning visual recognition, autonomous driving, biology, and healthcare. He aims to develop fundamental understandings and robust, widely applicable algorithms to tackle real-world challenges. He is particularly interested in learning from imperfect data, including limited, noisy, heterogeneous, distribution-shifting, and inaccessible data. His contributions have been recognized by several awards and honors, including the OSU Early Career Distinguished Scholar Award (2025) and CVPR Best Student Paper Award (2024). Before joining OSU in 2019, he was a Postdoctoral Associate at Cornell University (2018–2019), working with Kilian Weinberger and Mark Campbell. He earned his Ph.D. in Computer Science from the University of Southern California (2013–2018) under the supervision of Fei Sha.

Title: Neural-symbolic Software Auditing

Abstract: Software engineering is undergoing a major shift: while code generation has become increasingly automated, code review and audit remain stubbornly human-intensive. Code quality continues to be a persistent challenge, and developers often face the task of debugging or auditing code they did not write. LLMs hold promise for automating aspects of code analysis, yet they consistently fall short in auditing real-world repositories due to context limitations, hallucinations, and difficulty with repository-scale reasoning. In this talk, I will present our recent research that addresses these challenges. Specifically, RepoAudi is an autonomous, LLM-driven auditing agent designed for repository-level code analysis with high precision and efficiency. By mimicking expert auditors, it performs demand-driven, path-sensitive reasoning over control- and data-flow graphs, powered by abstraction, pointer tracking, and validation mechanisms. This approach has enabled RepoAudit to uncover hundreds of previously unknown bugs in mature software ecosystems, including the Linux kernel and OpenSSL. In order to extend the tool to audit binary executables that do not have any source code information, we developed a novel code-model training method and award-winning probabilistic program analysis approaches. These techniques enable RepoAudit to find dozens of vulnerabilities in real-world firmware by solely performing auditing. At the end, I will briefly discuss how to audit future software, namely, AI agents.

Bio: Xiangyu Zhang is a Samuel Conte professor at Purdue specializing in Software Engineering, AI red-teaming, and Cyber Forensics. His work involves developing techniques to detect bugs, including security vulnerabilities, in traditional software systems as well as AI models and systems, and to leverage AI techniques to perform software engineering and cybersecurity tasks. He has served as the Principal Investigator (PI) for numerous projects funded by organizations such as DARPA, IARPA, ONR, NSF, AirForce, and industry.

The Summit will feature a Doctoral Consortium on October 27 which will offer Ph.D. students opportunities to introduce their research to the broader research community, and receive mentorship, and constructive input. We are looking for individuals just before, or just after, their PhD proposal defense who are working in the area of human-centered AI, responsible AI, responsible computing, sustainability (environmental or community), or health.

For more information about speakers, visit the conference website.

Date: Wednesday, Nov. 12, 2025, at 12:00 p.m.

Location: Coda 230 (Google Maps link)

Lunch provided!

Title: Parametric Density Path Optimization

Abstract: We introduce Parametric Density Path Optimization (PDPO), a novel method for computing action-minimizing paths between probability densities. The core idea is to represent the target probability path as the pushforward of a reference density through a parametric map, transforming the original infinite-dimensional optimization over densities to a finite-dimensional one over the parameters of the map. We derive a static formulation of the dynamic problem of action minimization and propose cubic spline interpolation of the path in parameter space to solve the static problem. Theoretically, we establish an error bound of the action under proper assumptions on the regularity of the parameter path. Empirically, we find that using 3–5 control points of the spline interpolation suffices to accurately resolve both multimodal and high-dimensional problems. We demonstrate that PDPO can flexibly accommodate a wide range of potential terms, including those modeling obstacles, mean-field interactions, stochastic control, and higher-order dynamics. Our method outperforms existing state-of-the-art approaches in benchmark tasks, demonstrating superior computational efficiency and solution quality.

Bio: Sebastián Gutiérrez Hernández is currently a fifth-year CSE Ph.D. student in the Mathematics department at Georgia Tech. His main advisor is Dr. Haomin Zhou and he is co-advised by Dr. Peng Chen. Sebastián holds a bachelor's degree in Physics and Mathematics from the National Polytechnic Institute of Mexico and a Master's Degree in Applied Mathematics from the Research Center in Mathematics (CIMAT).
Sebastián's research focuses on computational methods at the intersection of optimal transport theory and scientific machine learning. His primary interest is extending the Wasserstein parametric framework to optimal transport problems with Lagrangian costs. As a secondary research direction, Sebastián leverages operator learning techniques to efficiently solve PDE-constrained optimization problems under uncertainty.

Sebastián is serving as the President of the SIAM Student Chapter at Georgia Tech since Fall 2025. If you are interested in attending or presenting in the chapter seminar series, please reach out to him.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

Title: Improving and Evaluating Streaming SVD for Large Scale Problems

Abstract: We study streaming singular value decomposition (SVD) for approximating large-scale SVD problems where data matrices are too large to store or recompute. Our focus is on Incremental SVD (iSVD), which applies hard-thresholding truncation after each streamed window. In contrast, Frequent Directions employs soft thresholding with provable but often loose bounds, and randomized methods such as SketchySVD construct one-pass sketches.  We propose algorithmic and theoretical improvements to iSVD and evaluate them experimentally.  

A “large-window” variant processes as many rows as memory allows, improving both accuracy and efficiency. Implemented in Skema, a high-performance C++ framework using Kokkos and PRIMME, this approach scales to massive datasets. Algorithmic contributions include warm-start iterative solvers reducing iterations by 30–50% and a reservoir-sampling–based convergence criterion for early stopping.

Theoretically, we outline a proof roadmap that when streaming rows in a random order (random shuffle), iSVD avoids pathological cases and achieves higher expected accuracy than Frequent Directions. Complexity analysis clarifies accuracy–cost trade-offs with SketchySVD, while experiments demonstrate that iSVD achieves several orders of magnitude higher accuracy at competitive runtime. As an application, we use iSVD to obtain better quality multigrid prolongators in Lattice Quantum Chromodynamics.

Bio: Andreas Stathopoulos is a Professor of Computer Science at William & Mary in Virginia, USA. He was awarded an NSF CISE Postdoctoral Fellowship after receiving his Ph.D. and M.S. in Computer Science from Vanderbilt University; he also completed a B.S. in Mathematics from the University of Athens in Greece. Dr. Stathopoulos’ research interests include numerical analysis and high performance computing; methods for large eigenvalue problems and linear systems of equations; and related applications from materials science and quantum chromodynamics. He co-developed PRIMME (Preconditioned Iterative MultiMethod Eigensolver), one of the foremost eigenvalue packages, several other significant software tools, and has published numerous journal articles and conference papers in computational sciences and applications.  He is a member of IEEE, IEEE Computer, and SIAM. Since 2010 he is an Associate Editor in SIAM Journal on Scientific Computing and from 2016-2022 served as its Section Editor for Software and HPC.

Speaker: Cory Hauck, applied mathematician at Oak Ridge National Laboratory
Date and Time: October 24, 2:00-3:00 p.m.
Location: Coda 114
Host: Qi Tang

Title: TBD

Abstract: TBD

Bio: TBD

September is Hunger Action Month and TSO is honored to serve as September Angels for Klemis Kitchen by hosting a food drive for Georgia Tech students on behalf of the College of Computing.  We invite you to participate and become a Klemis Kitchen angel by donating non-perishable food items for students in need.

Klemis Kitchen is Georgia Tech’s on-campus food pantry for students who are food under-resourced. This is a vital resource to allow those students to focus on their studies without the constant concern of food insecurity.  Klemis Kitchen collects shelf-stable items donated from food drives and other departments and clubs around campus, and they keep the pantry refrigerators and freezer stocked with leftover food recovered from the campus dining halls.

Date: Wednesday, Oct. 8, 2025, at 12:00 p.m.

Location: Coda, 9th Floor Atrium (Google Maps link)

Lunch provided!

Title: Design, development, and analysis of a compressible fluid dynamics solver capable of exceeding one quadrillion degrees of freedom

Abstract: This talk presents an optimized implementation of the recently proposed information geometric regularization (IGR) for unprecedented scale simulation of compressible fluid flows applied to multi-engine spacecraft boosters. The work improves upon state-of-the-art computational fluid dynamics (CFD) techniques in terms of computational cost, memory footprint, and energy-to-solution metrics. Unified memory on coupled CPU-GPU or APU platforms increases problem size with negligible overhead on NVIDIA and AMD architectures. Mixed half/single-precision storage and computation on well-conditioned numerics is used. The resulting solver is capable of simulating flow at 200 trillion grid points and 1 quadrillion degrees of freedom, exceeding the current record by a factor of 20. This talk describes the tools and languages used to develop fast and portable code that leverages coupled CPU-GPU and APU architectures and summarizes performance results that demonstrate real-world usability and meaningful improvement over the current state-of-the-art.

Bio: Ben Wilfong is a fourth-year CSE PhD student in Spencer Bryngelson’s computational physics group. His current research interests focus on high-performance computing for modeling and simulation applications on the world’s fastest supercomputers. He is a core developer for the Multi-Component Flow Code (MFC), an exascale compressible multiphase and multipysics flow solver.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

sign up and details

For more about CSE GSE, check out: https://cse.gatech.edu/cse-graduate-student-association

This event is set up differently from past CSE GSA hikes to accommodate increased interest. Three or four small groups meet at CODA and depart at 8:30 am, 10:00 am, 11:30 am, and/or 1:00 pm. These small groups will be assigned so that each group has students who can provide transportation to those who need it. Each group will also be assigned a group leader to select and lead the hike. 

Please provide accurate preferences for meeting times and cancel your RSVP if necessary so we can accommodate as many students as possible without last-minute cancellations. The sign-up is available here and CODA meetup times will be sent in the afternoon or evening of September 26th. Priority will be given in the order responses are received.

Thursday, Sept. 11 
6:00 - 8:00 p.m.
Tech Rec
John Lewis Student Center
Pizza to be provided

For more about CSE GSE, check out: https://cse.gatech.edu/cse-graduate-student-association

Date: Wednesday, Sept. 17, 2025, at 12:00 p.m.

Location: Coda, Room 114 (Google Maps link)

Lunch provided!

Title: Back to Bits: Extending Shannon’s Communication Performance Framework to Computing

Abstract: This presentation proposes a novel measurement unit for computing performance grounded in information theory. Modern computing systems are increasingly diverse, supporting low-precision formats, hardware specialization, and emerging paradigms such as analog, quantum, and reversible logic. Traditional metrics like floating-point operation (flop) counts no longer accurately capture this complexity. Instead, I frame computing as the transformation of information through a channel and define performance in terms of the mutual information between a system’s inputs and outputs. This approach measures not just the quantity of data processed, but the amount of meaningful information encoded, manipulated, and retained through computation. The framework provides a principled, implementation-agnostic foundation for evaluating performance.

Bio: Max Hawkins is a CSE PhD student interested in computer performance evaluation, sustainable data center operation, and adapting hardware systems to each running workload. His previous experience in radio astronomy, analog hardware, and data center design compelled him to explore a more general approach to application-agnostic computing performance evaluation. This presentation covers the current state of his work on this topic. He is advised by Richard Vuduc and Spencer Bryngelson.

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.

• Website Management Options Discussion
  (CampusPress/OIT Hosting/CoC VMs)
• Website Management Q&A
• General Q&A

Sign-in starts one hour before each race for all age groups.

• Youth (ages 7-12) – sign-in starts at 7:30 a.m. | race starts at 8:30 a.m.
• Older Youth (ages 13-18) - sign-in starts at 8:30 a.m. | race starts at 9:30 a.m.
• Open and Master's - sign-in starts at 9:30 a.m. | race starts at 10:30 a.m.

Fees: FREE for College of Computing faculty, staff, and students. Present your Buzz Card or GTID at sign-in. | $5.00 for (non-CoC) GT faculty, staff, and students.

*Check back soon for the registration link.

Panelists include:

• Kim Toatley, Interim Executive Vice President for Administration and Finance, Chief Business Officer 
• Ron Johnson, Chief of Staff and Senior Vice President of Strategic Initiatives,  Professor of the Practice, Industrial and Systems Engineering  
   

Time: 10 to 11 a.m.
Location: The Bill Moore Student Success Center, Clary Theater 

In-person attendance is now closed as we have reached our maximum room capacity.

If you did not RSVP to attend in person, join us virtually via Zoom. 

As always, we are reserving time for an open Q&A, and you are encouraged to take a minute to submit your questions in advance to evpaf@gatech.edu.

Title: Mapping Tissue Organization and Development Across Space and Time

Abstract: The development of a multicellular organism from a single founder cell is the result of a complex process involving growth, death, and differentiation of cells.  Recent technologies help quantify this process: spatial sequencing technologies measure spatial variation in cell types and gene expression across a tissue slice, while new genome editing technologies enable the tracking of cellular lineages during tissue development.  In this talk, I will describe to reconstruct development processes using data from these technologies.  First, I will present algorithms to align spatiotemporal transcriptomics data from multiple timepoints during the development of an organism.  These algorithms use optimal transport to infer ancestral relationships between cell types and to estimate cellular growth rates.  Next, I will present an algorithm to infer a cell differentiation map – a directed graph describing transitions between cell types – from single-cell lineage tracing data.  This algorithm models unmeasured progenitor cell types and infers a cell differentiation map that balances the trade-off between the complexity of the map and the number of unobserved cell type transitions.  These algorithms give insight into the gene expression programs that govern the growth and differentiation of cells over space and time.

Bio: Ben Raphael is the Graduate Class of 1991 Professor of Computer Science at Princeton University, and an affiliate faculty member at the Columbia University Irving Institute for Cancer Dynamics, the New York Genome Center, and the Rutgers Cancer Institute of New Jersey.  His research in computational biology focuses on single-cell and spatial DNA/RNA sequencing, cancer evolution, lineage tracing, network/pathway analysis of genomic variation, and structural variation in human and cancer genomes.  His group’s algorithms have been used in multiple large-scale cancer genomics projects including The Cancer Genome Atlas (TCGA), International Cancer Genome Consortium (ICGC), and Human Tumor Atlas Network (HTAN).  He is Fellow of the Association for Computing Machinery (ACM) and the International Society for Computational Biology (ISCB).  He is the recipient of the 2021 Innovator Award from the ISCB, the Alfred P. Sloan Research Fellowship, the NSF CAREER award, and a Career Award at the Scientific Interface from the Burroughs Wellcome Fund.