Title: Building Efficient and Scalable Machine Learning Systems

Time: 10:00pm 

Venue: CB 328

Speaker(s): Dr. Qinghao Hu

Remark(s): 

Abstract

The rapid evolution of foundation models is increasingly bottlenecked by a widening gap between algorithmic demands and system efficiency. As model scale and context lengths explode, infrastructure efficiency plateaus. Addressing these challenges requires a full-stack rethinking of machine learning systems. In this talk, I will present a research framework centered on algorithm–system co-design to push the efficiency frontier across the ML lifecycle. I first demonstrate how system-level support for algorithm advancement can drastically reduce the cost of large-scale hyperparameter exploration. Next, I tackle the long-tailed execution bottlenecks in post-training reinforcement learning, introducing a co-designed system approach that delivers substantial efficiency gains while preserving on-policy RL training. I also introduce system designs that enable vision–language models to scale to million-token contexts by resolving fundamental memory and communication constraints. I conclude by discussing the future of system support for agentic models at scale.

About the speaker

Qinghao Hu is a Postdoctoral Associate at the Massachusetts Institute of Technology, advised by Professor Song Han. His research focuses on efficient machine learning systems, spanning datacenter scheduling, distributed training, reinforcement learning, and model serving. His work has been recognized with the ASPLOS Distinguished Paper Award and the WAIC Best Paper Award. He is a recipient of the Google Ph.D. Fellowship, the ML and Systems Rising Stars Award, and the Best Ph.D. Thesis Award. His research has been featured in MIT News, NTU News, and the USENIX ;login: newsletter, and his open-source projects have attracted more than 6,000 GitHub stars. He received his Ph.D. from Nanyang Technological University and was previously a visiting scholar at ETH Zürich.

Title: Advancing Exploration in Reinforcement Learning

Time: 02:00pm 

Venue: CB 328

Speaker(s): Prof. Leong Hou U

Remark(s): 

Abstract

Exploration remains a key barrier to deploying reinforcement learning in realistic embodied settings, where agents must act under high-dimensional visual observations, sparse and delayed rewards, and often overactuated control interfaces. This talk presents a line of research that makes exploration more practical and scalable by progressively introducing structure into both representation and intrinsic motivation. We first revisit metric-based intrinsic bonuses and propose an effective discrepancy metric with adaptive scaling to improve robustness on hard exploration benchmarks. We then move beyond raw novelty by learning compact representations in a behavioral metric space and rewarding value-diverse, behaviorally distinct trajectories for scalable exploration in high-dimensional environments. To address long-horizon embodied tasks, we introduce latent “foresight” via diffusion-based self-prediction and a latent-space exploration reward, demonstrating gains in navigation/manipulation and real-world indoor deployment. Finally, for overactuated musculoskeletal control, we discover disentangled synergy patterns and learn policies entirely in a synergy-aware latent action space, improving efficiency and generalization.

About the speaker

Leong Hou U is currently an Associate Professor in the Department of Computer and Information Science at the University of Macau, Director of the Data Science Center. His research focuses on interdisciplinary areas at the intersection of artificial intelligence and data engineering, including traffic data optimization, spatiotemporal databases, large-scale data visualization, graph neural networks, and reinforcement learning. His team has published over 80 papers in leading journals and conferences such as SIGMOD, VLDB, ICDE, NeurIPS, AAAI, ICLR, IJCAI, and KDD. In recent years, the team has led and participated in multiple national and regional key R&D projects, including the National Key R&D Program on efficient integration and dynamic cognition technologies for urban public services, the Macau Science and Technology Development Fund key project on collaborative intelligence–driven autonomous driving, and a 2024 project on urban traffic perception fusion and intelligent reasoning that received the Second Prize of the Science and Technology Invention Award. He is also actively engaged in the international research community, serving in program and organizing committees for major conferences such as BigData, IJCAI, ICDE, DASFAA, and PAKDD, and has been a committee member of the China Association of Young Scientists (Information and Electronic Science) and the Urban Planning Committee of the Macao SAR Government since 2020, promoting the integration of scientific research with urban development policy.

Title: Using Optimal Transport To Mitigate Unfair Predictions and Quantify Counterfactual Fairness

Time: 11:00am 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Arthur Charpentier

Remark(s): 

Abstract

Many industries are heavily reliant on predictions of risks based on characteristics of potential customers. Although the use of said models is common, researchers have long pointed out that such practices perpetuate discrimination based on sensitive features such as gender or race. Given that such discrimination can often be attributed to historical data biases, an elimination or at least mitigation, is desirable. With the shift from more traditional models to machine-learning based predictions, calls for greater mitigation have grown anew, as simply excluding sensitive variables in the pricing process can be shown to be ineffective.

In the first part of this seminar, we propose to mitigate possible discrimination (related to so call group fairness, related to discrepancies in score distributions) through the use of Wasserstein barycenters instead of simple scaling. To demonstrate the effects and effectiveness of the approach we employ it on real data and discuss its implications.

In the second part, we will focus on another aspect of discrimination usually called counterfactual fairness, where the goal is to quantify a potential discrimination if that person had not been Black or if that person had not been a woman. The standard approach, called ceteris paribus (everything remains unchanged) is not sufficient to take into account indirect discrimination, and therefore, we consider a mutates mutants approach based on optimal transport. With multiple features, optimal transport becomes more challenging and we suggest a sequential approach based on probabilistic graphical models

About the speaker

Professor Arthur Charpentier
Department of Mathematics
University of Quebec at Montreal

Title: Accommodating Comprehensive Foundation Model Services over Heterogeneous Computational Resources

Time: 02:00pm 

Venue: CB 328

Speaker(s): Prof. Binhang Yua

Remark(s): 

Abstract

Deploying foundation model services is crucial to contemporary AI applications. We focus on deploying such services in heterogeneous, potentially decentralized settings to mitigate the substantial costs typically associated with centralized data centers. Our work relies on carefully designed scheduling algorithms and integrated system optimizations to fully unleash the potential of heterogeneous computational power across comprehensive serving paradigms, including data preparation pipelines, large-scale pretraining, reinforcement learning-based alignment, and agentic inference deployment.

About the speaker

Binhang YUAN is an Assistant Professor at the Department of Computer Science and Engineering (CSE), the Hong Kong University of Science and Technology (HKUST) since 2023. He received his Ph.D. and master's degrees from Rice University and his bachelor's degree from Fudan University. Before joining HKUST, he was a Postdoc at the Swiss Federal Institute of Technology Zurich (ETH Zurich). His main research interests are in distributed, decentralized, and heterogeneous machine learning systems for foundation models.

Title: Quantum Recursive Programming: Abstraction, Verification, and Implementation

Time: 10:00am 

Venue: CB 308

Speaker(s): Dr. Zhicheng Zhang

Remark(s): 

Abstract

Recursion is a fundamental concept in computer science, allowing complex algorithms to be expressed as compact, elegant, and modular programs. In quantum computing, recursion is appealing for the same reasons, but it also interacts in subtle ways with uniquely quantum features such as superposition and entanglement. Today, many quantum algorithms are still presented as low-level circuits, making them difficult to analyse, verify, and scale. In this talk, I will introduce quantum recursive programming, a high-level framework that supports modular programming of quantum algorithms. I will discuss two complementary perspectives on quantum recursive programs: a proof system for the formal verification of their correctness, and a framework for their efficient implementation. Together, these results show how quantum recursive programming can bring together modularity, correctness, and efficiency.

About the speaker

Zhicheng Zhang is a postdoctoral research fellow at the University of Technology Sydney. He earned his PhD under the supervision of Mingsheng Ying, Zhengfeng Ji, and Sanjiang Li. His research spans the theory of quantum computing, with a particular focus on the design of efficient quantum algorithms and the quantum software foundations needed for their high-level programming and reliable implementation.

Title: A Billion Medical Devices

Time: 10:30am 

Venue: CB 308

Speaker(s): Prof. Mayank Goel

Remark(s): 

Abstract

As we live longer, we are also living with more diseases. The need to identify illness symptoms and manage them has been increasing rapidly. Our personal devices have a role to play in helping us take care of ourselves outside a doctor’s clinic or hospital. Technology’s role in healthcare is already quite ubiquitous in the form of step counters and heart rate monitors. However, we can go far beyond these coarse measures. I will provide an overview of our efforts at building real-time machine-learning systems that measure depression symptoms, fatigue, sleep quality, and hyperactivity. While useful, these machine learning systems will probably never be perfectly accurate. The sensed information will always be noisy. Moreover, the user won’t know how to interpret and use measured information. I will present our ideas on how to make the inferred information actionable to the patients, caretakers, and doctors. I will also talk about the role of noisy machine learning systems and how a user can counter the system’s inherent error and uncertainty. Ultimately, we aim to build systems that help us manage our health without requiring perfectly accurate inferences.

About the speaker

Mayank is an Associate Professor in the Software and Societal Systems Department (S3D) and Human-Computer Interaction Institute (HCII) in the School of Computer Science at CMU. He leads the Smart Sensing for Humans (SMASH) Lab at CMU. His group develops new, practical, and deployable sensing and machine-learning solutions to solve problems in healthcare and reduce usage barriers for under-resourced populations. He regularly collaborates with mechanical and biomedical engineers, doctors, nurses, community health workers, and patients and their caregivers worldwide. His systems are currently used by hundreds of patients and deployed in several clinics and hospitals around the world.

Title: F.A.C.U.L.: The World’s First FPS AI Companion that Understands Human Language.

Time: 02:00pm 

Venue: HW 312

Speaker(s): Dr. Elvis Liu

Remark(s): 

Abstract

Arena Breakout is the first FPS game that allows players to interact with AI companions through natural language. Traditionally, FPS games restrict communication to simple commands like “attack” or “follow,” due to the limitations of existing input methods such as hotkeys and command wheels. These commands lack specificity, hindering players from giving complex tactical instructions like “clear the second floor” or “take cover behind that tree”.

Arena Breakout enables players to interact with an intelligent companion called F.A.C.U.L. using natural language. F.A.C.U.L. can understand complex instructions, provide feedback, and perform a series of tasks. Moreover, it can identify thousands of in-game objects, including buildings, vehicles, and collectable items, and accurately distinguish different colors and materials.

This revolutionary feature integrates advanced generative AI technologies, including voice input, LLMs, real-time text-to-speech, and image description, creating the most immersive experience for players and allowing them to work with human-like AI. F.A.C.U.L. was first presented at GDC 2025, the most important conference of the gaming industry, and later published at AAAI 2026.

About the speaker

Dr. Elvis Liu is the Head of AI at MoreFun Studios, Tencent, leading R&D in game AI, generative AI, physics simulation, and distributed virtual environments. He spearheaded the deep-reinforcement-learning sparring agents for Naruto Mobile, earning the Tencent Technology Breakthrough Award 2024, and created F.A.C.U.L., the first FPS AI companion enabling natural-language interaction with players.

Dr. Liu has over 20 years of experience spanning parallel and distributed systems, computer simulations, AI, computer graphics, and high‑performance computing. His contributions to these research areas have been acknowledged by invitations to serve on the Programme Committee of the key conferences in these research fields, including ACM SIGSIM-PADS, ACM/IEEE DS-RT, and Winter Simulation Conference. He is currently an Associate Editor of IEEE Transactions on Games and was named an Overseas High-Level Talent by the Chinese government in 2018.

Previously, he was an Assistant Professor at Nanyang Technological University in Singapore and an IRCSET Fellow in Ireland, where his IBM research on parallel simulation for exascale supercomputers was selected for ACM’s 19th Annual Best of Computing list, a collection of the best articles published in computing.

Title: Edge Intelligence: Multi-Modality Sensing to Revolutionize Chronic Diseases Management

Time: 10:30am 

Venue: HW312, Haking Wong Building

Speaker(s): Dr. Qian Zhang

Remark(s): 

Abstract

Rapid aging has led to an increase in the number of chronic disease patients, rising social medical costs, and a decline in the quality of life of patients. Under today's hospital-centered service model, patients are already seriously ill when diagnosed, and their subsequent prognosis is quite uncertain. The research attempts to transform the service model into a lifecycle management model, using IoT sensors and AI algorithm design for early diagnosis of diseases, and will also provide continuous health assessment and rehabilitation guidance to reduce the cost of medical services.

Dr. Qian Zhang will share their recent work on chronic disease assessment and intervention supported by intelligent sensing (multiple sensing modality will be leveraged) and AI algorithm design. She will use cuffless continuous blood pressure monitoring (leveraging mmWave and fiber optics), as well as COPD assessment and training (leveraging audio, IMU, as well as depth camera), as example for the sharing.

About the speaker

Dr. Qian Zhang joined the Hong Kong University of Science and Technology (HKUST) in September 2005. She is currently the Head of the Division of Integrative Systems and Design (ISD), Tencent Professor of Engineering, and Chair Professor of the Department of Computer Science and Engineering (CSE) at HKUST. She is also the Director of the Microsoft Research Asia-HKUST Joint Laboratory, the Co-Director of the Huawei-HKUST Innovation Laboratory. Prior to this, she worked at Microsoft Research Asia from July 1999 as a research manager in the Wireless and Networking Group. Dr. Zhang has published more than 500 refereed papers in leading international journals and major conferences. She is the inventor of more than 50 international patents. Her current research interests include the Intelligent Internet of Things (AIoT), smart health, mobile computing and sensing, wireless networks, and network security.

Dr. Qian Zhang is an IEEE Fellow and a Fellow of the Hong Kong Academy of Engineering (HKAE). Dr. Zhang served as the Editor-in-Chief of IEEE Transactions on Mobile Computing from 2020 to 2022. She is currently a member of the IEEE Infocom Steering Committee.

Title: Guiding with Search: Memory-Aware Abstraction for Scalable Vulnerability Detection

Time: 10:30am 

Venue: CB 308

Speaker(s): Prof. Heqing Huang

Remark(s): 

Abstract

Along with software development, system reliability and robustness have become major concerns. However, due to the complex program semantics, efficiently detecting these vulnerabilities remains challenging. This talk introduces a new perspective, search-space guided analysis with adapted memory organization, to significantly improve the performance of vulnerability detection. Unlike existing efforts that focus solely on the code itself, our approach integrates abstraction with runtime memory as both an oracle and a guidance for better vulnerability detection. The related work has received the ASPLOS Best Paper award and the Google Research Paper award. The artifact has also been successfully integrated into the most widely used compiler infrastructure, LLVM.

About the speaker

Heqing Huang is an assistant professor at the Department of Computer Science at City University of Hong Kong. His research focuses on software security, especially leveraging program analysis techniques to ensure software security rigorously. Specifically, his research takes advantage of both static and dynamic program analysis techniques as complements to address deficiency problems in existing vulnerability detection methods, such as fuzzing, symbolic analysis, and memory sanitization. On the other hand, he also aims to demonstrate the practicalness of the general research methods on specific application scenarios, e.g., Android and Linux kernels. His research has received the ASPLOS Best Paper and Google Research Paper awards. He also served on the CCS program committee and as a reviewer for TDSC, TOSEM, and TSE.