Ken Kennedy Institute AI Seminar Series | Fridays 12:00–12:50 PM Central
The Rice Computer Science department, Graduate Student Association (CS GSA), and Ken Kennedy Institute will host a weekly AI Seminar on Fridays from 12:00–12:50 PM at Rice University. This event is open to Rice University graduate students; no RSVP required. Join us each week at noon for a light lunch before the talk begins at 12:15 PM. See the list of seminars below.
| Date | Location | Seminar Details (Fall 2025) |
|---|---|---|
| 9/5 | DH 1070 | Speaker: M.S.E. Gregory Holste (Ph.D. Candidate, UT Austin) Talk: Deep Learning for Automated Echocardiography Interpretation In this talk, I will trace the evolution of AI-enabled echocardiography interpretation from specialized, single-task models to comprehensive automated systems through my own PhD research. This talk will begin with foundational work on self-supervised learning and aortic stenosis detection from a single view of the heart, then culminate with PanEcho, an end-to-end system for multi-view echocardiogram interpretation that mirrors a real-world clinical workflow. I will conclude by discussing what is needed to translate these innovations into clinical practice and how these advances can democratize access to cardiovascular healthcare worldwide. View full talk details on the event calendar here. |
| 9/12 | DH 1070 | Speaker: Prof. Tomer Galanti (Assistant Professor, Texas A&M University) Talk: Self-Supervised Contrastive Learning ≈ Supervised Contrastive Learning In this talk, I will present theoretical and empirical results showing that self-supervised contrastive learning is, in a precise sense, approximately equivalent to a supervised contrastive objective. I will describe a simple, model-agnostic relationship linking self-supervised and supervised contrastive losses, and demonstrate the strong alignment of their learned representations. Furthermore, we extend the notion of neural collapse to the self-supervised setting and introduce a new geometric measure that captures this property. Leveraging this, we derive a novel bound that directly connects representation geometry to downstream transfer performance, offering a principled explanation of why contrastive learning generalizes so effectively. View full talk details on the event calendar here. |
| 9/19 | DH 1070 | Speaker: Dr. Chen Luo (Sr. Applied Scientist, Amazon Research) Talk: From Search to Conversational Shopping at Amazon via Generative Models In this talk, I will share our multi years journey of re-imagining online shopping at Amazon—transforming it from a product search problem into a conversational shopping experience powered by generative models. View full talk details on the event calendar here. View full talk details on the event calendar here. |
| 9/26 | DH 1070 | Speaker: Prof. Shibbir Ahmed (Assistant Professor, Texas State University) Talk: Engineering Trust in AI: Design by Contract for Reliable Predictions In this talk, the speaker will present techniques for improving the reliability of AI software. First, the talk will introduce data preconditions, computed from the model structure and trained parameters, which determine when outputs from a DL model should not be trusted. Next, the speaker will emphasize a contract layer in the deep learning library that intercepts API calls to check contracts, helping detect and fix bugs often ignored by DL libraries. An example will be shown in the context of safety specifications for clinical AI models. Finally, the talk will conclude with a discussion of open challenges at the intersection of Software Engineering and AI, and what trustworthy AI could mean for real-world deployments. View full talk details on the event calendar here. |
| 10/3 | DH 1070 | Speaker: Prof. Kuan-Hao Huang (Assistant Professor, Texas A&M University) Talk: Toward Robust and Reliable Large Language Models In this talk, I will explore position-related robustness issues across two aspects of LLMs: pure text-based LLMs and multimodal LLMs. Specifically, I will first introduce how position bias can hurt the understanding capabilities of LLMs and present a training-free solution to address this issue. Next, I will discuss position bias in the multimodal setting and introduce a simple plug-in module to enhance robustness in multimodal understanding. Finally, I will discuss some potential future research directions. View full talk details on the event calendar here. |
| 10/10 | DH 1070 | Speaker: Prof. Aniruddha Bora (Assistant Professor, Texas State University) Talk: Mathematically Diverse Multiscale Neural Operators: From Bias Correction to Inverse Forcings Most real-world phenomena are governed by high-dimensional, nonlinear dynamical systems. This talk will address two complementary tasks: (i) the forward problem (improving predictions when simulators exhibit structural bias or under-resolution) and (ii) the inverse problem (given temperature response fields, infer the imposed spatial forcing). View full talk details on the event calendar here. |
| 10/17 | DH 1070 | Speaker: Prof. Jingchao Ni (Assistant Professor, University of Houston) Talk: Cross-Modal Knowledge Transfer in Time Series via Multimodal Views In this talk, I will provide an overview of recent developments in large foundation models for time series, highlighting frameworks for transferring knowledge from other modalities to time series. I will then delve into the emerging area of cross-modal knowledge transfer via multimodal views (MMVs) of time series, discussing (1) the generation of MMVs (e.g., linguistic and visual views) of time series; (2) methods for modeling time series via MMVs; and (3) the integration of MMVs with multimodal models. This talk aims to review state-of-the-art AI techniques for time series, highlight unique challenges, and share our recent findings in this promising area. View full talk details on the event calendar here. |
| 10/24 | DH 1070 | Speaker: Prof. Sen Lin (Assistant Professor, University of Houston) Talk: Theory toward Demystifying Continual Learning Continual Learning (CL) has attracted significant attention in recent years because of its importance and broad applications in enabling lifelong learning capabilities without forgetting old knowledge. However, most studies in this area are empirical, and the theoretical understanding of CL is still in early stage. In this talk, I will share our recent studies for addressing this gap, which not only build the theoretical foundations toward demystifying CL but also inspire new algorithm design in a principled way. View full talk details on the event calendar here. |
| 10/31 | DH 1070 | Speaker: Prof. Zhaozhuo Xu (Assistant Professor, Stevens Institute of Technology) Talk: Behavior-Aware Data Valuation for LLMs at Scale Large Language Models (LLMs) depend on massive datasets whose quality and influence remain largely opaque. Data valuation offers principled methods to quantify how training data contributes to model performance and behavior. Yet, scaling classical approaches such as influence functions to trillion-token corpora continues to be a major challenge. This talk introduces recent advances that address this gap, including the linearized influence kernel, a new and efficient metric that extends to LLMs with billion-scale parameters. View full talk details on the event calendar here. |
| 11/7 | DH 1070 | Speaker: Prof. Beidi Chen (Assistant Professor, Carnegie Mellon University) Talk: Rethinking Test-Time Scaling Laws Exploring new dimensions in scaling has consistently driven progress in generative AI—from larger models and datasets to extended context lengths, and now toward test-time scaling (TTS). In this talk, we first rethink test-time scaling laws from a practical efficiency perspective, demonstrating that the benefits of smaller models are often overstated when considering memory bottlenecks introduced from inference-time strategies (e.g., Best-of-N, long CoTs). View full talk details on the event calendar here. |
| 11/14 | DH 1070 | Speaker: Prof. Golnaz Habibi (Assistant Professor, University of Oklahoma) Talk: Scalable Collaboration in Multi-Agent Robotic Systems: From Perception to Decision-Making Autonomous multi-agent systems are transforming the way robots perceive, decide, and act in dynamic and uncertain environments. As robotic teams grow in size and complexity, enabling effective cooperation requires new approaches to distributed perception, communication, and decision-making. In this talk, I will present recent advances in scalable algorithms that allow robotic agents to collaboratively interpret their surroundings, exchange information efficiently, and make resilient decisions. View full talk details on the event calendar here. |
| 11/21 | DH 1070 | Speaker: Prof. Jie Cao (Assistant Professor, University of Oklahoma) Talk: Decoding Classroom Dialogue: Discourse Analysis in STEM Education This talk explores how recent advances in language modeling can enhance Educational AI by supporting dialogue-driven learning in STEM classrooms. We will introduce our studies on providing pedagogical feedback to educators using multiple theoretical frameworks on discourse analysis. View full talk details on the event calendar here. |
| 12/5 | DH 1070 | Speaker: Prof. Xuan Lu (Assistant Professor, University of Arizona) Talk: Towards Sustainable Remote Work: A Practice of Human-Centered Data Science In this talk, I will introduce a comprehensive human-centered data science approach (especially featuring machine learning and causal inference) to analyzing and improving the sustainability of remote workers and teams. In particular, the presented work will focus primarily on the open-source software development community and analyze the factors that affect the sustainability of virtual teams under shock and promote the long-term contribution of individual developers. View full talk details on the event calendar here. |
To see a list of all upcoming events related to the Ken Kennedy Institute, visit https://kenkennedy.rice.edu/events.
For questions, please contact at sj157@rice.edu.
-
See the spring 2025 schedule below:
-
Date Location Seminar Details (Spring 2025) 2/27 DH 1046 Speaker: Prof. Na Zou, PhD (Assistant Professor, University of Houston)
Talk: Exploring and Exploiting Shortcuts in AI/ML: Algorithms, Challenges and Solutions
For speaker bio and more information, please check the event page: Here
AI/ML algorithms have made significant advancements and are extensively used in critical applications such as employment, personalized medicine, and more. Despite the success, mitigating shortcut features and spurious correlation in AI/ML remains a significant challenge. This talk will provide a computational perspective on shortcuts and spurious correlation in AI/ML, encompassing the measurement, detection, and mitigation of shortcuts and spurious correlation to address diverse challenges throughout the AI/ML life cycle. The speaker will first introduce real-world examples, fundamental concepts and the existing work. The speaker will also highlight her ongoing research across three key stages in AI/ML: enhancing data quality, refining algorithmic design, and optimizing model deployment.3/6 DH 1070 Speaker: Prof. Jessica Ouyang, PhD (Assistant Professor, UT Dallas)
Talk: Generating Scientific Literature Reviews
For speaker bio and more information, please check the event page: Here
As researchers, we spend a lot of time interacting with literature reviews: we read them when we want to quickly get up to speed in a new research area, and we write them when we need to ground our own research contributions in the landscape of existing work. In this talk, I will discuss my lab's work on developing neural approaches to generate scientific literature reviews. I will begin by highlighting our contributions in the task of citation generation in the context of a larger literature review paragraph, including generation at the sub- and multi-sentence levels, enforcing coherence with the surrounding paragraph, and retrieving grounding sentences from cited papers. Then, I will introduce our work on full-length literature review generation and describe a detailed user study on the capabilities of current state-of-the-art language models on this task. Finally, I will conclude with a discussion of current challenges in literature review generation, as well as the ethical considerations that arise when automating part of the scientific writing process.3/13 O'Connor 5th Floor Speaker: Prof. Mauricio Araya Polo, PhD (Adjunct Professor, Rice; Senior R&D Manager, Total Energies)
Talk: Addressing Geophysical Problems with Scientific ML
For speaker bio and more information, please check the event page: Here
Advances in machine learning (ML) are open new avenues for complex geophysical problems. This seminar explores how scientific ML techniques, in particular DL-driven inversion, can enhance traditional geophysical methods. We illustrate their application to seismic inversion, subsurface characterization, and CO2 monitoring through examples. Attendees will gain insights into current methodologies and future directions in integrating scientific ML with geosciences.3/20 N/A No Seminar — Spring Break 3/27 DH 1070 Speaker: Prof. David Harwath, PhD (Assistant Professor, UT Austin)
Talk: Speech Generation and Sound Understanding in Era of Large Language Models
For speaker bio and more information, please check the event page: Here
Transformer-based large language models (LLMs) have rapidly risen to dominance in the NLP field. One of the most exciting developments in this line of research is the finding that LLMs can be easily extended to handle multimodal inputs, such as vision or speech, via tokenization and concatenation with natural language inputs. In this talk, I will discuss several of my group's recent research directions into expanding the capabilities of multimodal LLMs to process speech and spatial audio signals4/3 O'Connor 5th Floor Speaker: Prof. Sunyang Fu, PhD (Assistant Professor, UT Health Houston)
Talk: Advancing Aging Research with Natural Language Processing and Electronic Health Records: Opportunities, Challenges, and Future Directions
For speaker bio and more information, please check the event page: Here
Routinely collected real-world data, such as electronic health records (EHRs), serve as a valuable resource for aging research and geriatric medicine. However, the unstructured nature of these data presents significant challenges. To address this, natural language processing (NLP) techniques have been developed and applied to extract and standardize information from unstructured clinical text. This talk will provide an overview of NLP, highlighting its relevance to EHR data and its applications in aging research and geriatric medicine. I will also discuss some of the key challenges to adopting NLP in clinical research and practice and future research opportunities.4/10 DH 1070 Speaker: Jyoti Anand (Director of Data Science, Walmart)
Talk: Gen AI in Retail Technology
As the director of Walmart Data Science team, Jyoti has been integrating ML solutions with business functions to create the next generation of AI-powered capabilities. Her rich industry experience in Allegis, FDA, and Walmart has given her deep insight in Forecasting, LLMs, Deep Learning models.4/17 DH 1070 Speaker: Prof. Tianlong Chen, PhD (Assistant Professor, University of North Carolina at Chapel Hill)
Talk: Multimodal LLM Agents: Profile, Memory, Planning, and Actions
For speaker bio and more information, please check the event page: Here
Large Language Model (LLM) agents are emerging as a powerful paradigm for enabling intelligent decision-making, planning, and interaction across a wide range of real-world applications. In this talk, I will introduce the concept of LLM agents and explore recent advances in designing more powerful, scalable, and efficient agent systems. I will cover core challenges in multi-agent collaboration, such as communication strategies, agent specialization, and memory management, and present recent innovations including Self-MoA, Graph-of-Agents (GoA), and Symbolic-MoE. I will also discuss retrieval-augmented generation (RAG) extensions, such as Chain-of-Agents and Reflective Memory Management, to improve reasoning over long contexts. Additionally, I will share recent progress in using LLM agents for embodied decision-making in robotics and industrial settings, as well as novel use cases in scientific discovery and judgment modeling. Through a series of real-world case studies and prototypes, this talk highlights both the promises and open questions in building next-generation LLM agent systems.4/24 DH 1070 Speaker: Prof. Cheng Zhang, PhD (Assistant Professor, Texas A&M University)
Talk: See More From Less: Generative Human Modeling from Sparse Visual Cues
For speaker bio and more information, please check the event page: Here
Digital human reconstruction plays a critical role in many applications that benefit society, such as AR/VR, human-computer interaction, social robotics, and clinical diagnosis. However, generating accurate and detailed 3D humans remains challenging due to the scarcity of 3D data or multi-view inputs and the difficulty of capturing geometry and textures in uncontrolled environments. In this talk, I will discuss my recent work on steering generative models to reconstruct and manipulate 3D humans from sparse, or even single, visual inputs. The key idea is to combine generative modeling with human priors and appearance cues.
