Embedded Files
AICIT 2026
SPEAKERS
SPEAKERS
BRIEF INTRODUCTION
BRIEF INTRODUCTION
Dr. Huazi Zhang
Title:
Abstract:
Bio:
Prof. Eli Chien
Title: Beyond Public States: Tightening Differential Privacy Bounds in Machine Learning via Hidden-State Analysis
Abstract: Differential Privacy (DP) has emerged as the gold standard for privacy-preserving machine learning, seeing widespread adoption by government agencies and leading technology companies. The primary workhorse for guaranteeing DP in these models is Differentially Private Stochastic Gradient Descent (DP-SGD). However, standard privacy accounting for DP-SGD relies on composition theorems that aggregate the privacy loss induced at every update step. This implicitly assumes a "public-state" scenario, analyzing the privacy loss as if all intermediate model updates are released. Because only the final model is typically released in practice, this public-state assumption leads to overly pessimistic, divergent privacy bounds as the number of iterations grows.
In this talk, we introduce a hidden-state DP analysis to bridge this gap. First, we demonstrate how to conduct hidden-state analysis for non-smooth, non-convex losses in first-order optimization. This approach yields the tightest known hidden-state DP bound to date and resolves key open questions raised by Ye & Shokri (2022) and Altschuler & Talwar (2022). Finally, we extend this framework to zeroth-order optimization, addressing complex scenarios where traditional shifted-divergence analyses no longer apply.
Bio: Eli Chien is an Assistant Professor at National Taiwan University (Electrical Engineering, AI-CoRE) with the Yushan Young Fellow. He was a visiting researcher at Google Research (hosted by Peter Kairouz) and a Postdoctoral Fellow at the Georgia Institute of Technology, working with Professor Pan Li. He obtained his Ph.D. from the University of Illinois, Urbana-Champaign, advised by Professor Olgica Milenkovic. His current research focuses on privacy in machine learning, including machine unlearning and differential privacy, as well as their applications to graph machine learning. His previous research centered on designing better graph neural networks with theoretical guarantees. His work has been primarily published in top-tier machine learning, data mining, and information theory venues, including ICML, NeurIPS, ICLR, Transactions on IT, KDD, TheWebConf, AISTATS, AAAI and more.
Prof. Chandra Nair
Title: The Capacity Region of Classes of Sums of Broadcast Channels
Abstract: We introduce the class of "primary-broadcast channels," a unifying framework that encompasses several previously studied models, such as less-noisy and semi-deterministic broadcast channels. The central result of this work is the characterization of the capacity region for the "sum" of channels within this class.
The derivation of this capacity region presents a notable technical challenge: for the cases considered, the result does not follow from the standard alignment of Marton's inner bound and the UVW outer bound. Indeed, we identify instances where these bounds do not match. Additionally, since traditional outer bounds for product channels are not applicable in this setting, a different approach is required. We demonstrate that the capacity region is instead established through the "auxiliary receiver" technique—an outer-bounding device that successfully matches Marton's inner bound for these sum-channels.
This is joint work with Amin Gohari and Liu Yi.
Bio: Chandra Nair is a Professor in the Department of Information Engineering at The Chinese University of Hong Kong (CUHK), where he also serves as the Programme Director of the Mathematics and Information Engineering (MIEG) program. He received his B.Tech. from IIT Madras and his Ph.D. from Stanford University, followed by a postdoctoral fellowship at Microsoft Research. His research focuses on network information theory, information inequalities, and combinatorial optimization. A Fellow of the IEEE, he is a recipient of the 2016 IEEE Information Theory Society Paper Award and was a plenary speaker at the 2021 IEEE International Symposium on Information Theory (ISIT). His work is characterized by developing novel techniques to resolve long-standing conjectures and establish the capacity regions of fundamental communication models.
Prof. Minming Li
Title: Strategic Learning Approach for Deploying UAV-provided Wireless Services
Abstract: Unmanned Aerial Vehicle (UAV) have emerged as a promising technique to rapidly provide wireless services to a group of mobile users simultaneously. The article aims to address a challenging issue that each user is selfish and may misreport his location or preference for changing the optimal UAV location to be close to himself. Using algorithmic game theory, we study how to determine the final location of a UAV in the 3D space, by ensuring all selfish users' truthfulness in reporting their locations for learning purpose. To minimize the social service cost in this UAV placement game, we design strategyproof mechanisms with the approximation ratios, when comparing to the social optimum. We also study the obnoxious UAV placement game to maximally keep their social utility, where each incumbent user may misreport his location to keep the UAV away from him. Moreover, we present the dual-preference UAV placement game by considering the coexistence of the two groups of users above, where users can misreport both their locations and preference types (favorable or obnoxious) towards the UAV. Finally, we extend the three games above to include multiple UAVs and design strategyproof mechanisms with provable approximation ratios.
Bio: Minming Li is a Professor in Department of Computer Science, City University of Hong Kong. He received his Ph. D. and B.E. degree in the Department of Computer Science and Technology at Tsinghua University in 2006 and 2002 respectively. His research interests include algorithmic game theory, combinatorial optimization and algorithm design and analysis for scheduling problems. He has been consistently working on the theoretical aspects of computer science with a wide scope of application background. The full publication can be found at https://dblp.org/pid/78/6881.html
He is serving as associate editors for a number of journals including JAAMAS, Journal of Scheduling, Journal of Artificial Intelligence Research and Journal of Combinatorial Optimization. He has also served as chairperson of ACM Hong Kong Chapter from 2016 to 2018. He was a recipient of Teaching Excellence Award given by City University of Hong Kong and Outstanding Supervisor Award. Currently he is ACM Distinguished Speaker.
Prof. Yunghsiang S. Han
Title: A Fast Decoding Algorithm for Generalized Reed-Solomon Codes and Alternant Codes
Abstract: In this paper, it is shown that the syndromes of generalized Reed- Solomon (GRS) codes and alternant codes can be characterized in terms of inverse fast Fourier transform, regardless of code definitions. Then, a fast decoding algorithm is proposed, which has a computational complexity of O(nlog(n−k)+(n−k)log2(n−k)) for all (n,k) GRS codes and (n,k) al- ternant codes. Particularly, this provides a new decoding method for Goppa codes, which is an important subclass of alternant codes. When decoding the binary Goppa code with a length of 8192 and a correction capability of 128, the new algorithm is nearly 10 times faster than traditional methods. The decoding algorithm is suitable for the McEliece cryptosystem, which is a candidate for post-quantum cryptography techniques.
Bio: Yunghsiang S. Han received B.Sc. and M.Sc. degrees in elec- trical engineering from the National Tsing Hua University, Taiwan, in 1984 and 1986, respectively, and a Ph.D. degree from the School of Computer and Information Science, Syracuse University, NY, in 1993. He was with Hua Fan College of Humanities and Technology, National Chi Nan Univer- sity, and National Taipei University, Taiwan. From August 2010 to January 2017, he was with the Department of Electrical Engineering at the National Taiwan University of Science and Technology. he was with the School of Electrical Engineering & Intelligentization at Dongguan University of Tech- nology, China, from February 2017 to February 2021. Now, he is with the Shenzhen Institute for Advanced Study at the University of Electronic Sci- ence and Technology of China. He is also a Chair Professor at National Taipei University since February 2015. Dr. Han’s research interests include error-control coding, wireless networks, and security. Dr. Han has been con- ducting state-of-the-art research in decoding error-correcting codes for more than twenty years. He first developed a sequential-type algorithm based on Algorithm A* from artificial intelligence. At the time, this algorithm drew a lot of attention since it was the most efficient maximum-likelihood decoding algorithm for binary linear block codes. Dr. Han has also successfully applied coding theory to wireless sensor networks. He has published several highly cited works on wireless sensor networks, such as random key pre-distribution schemes, and serves as the editor of several international journals. Dr. Han was the winner of the Syracuse University Doctoral Prize in 1994 and a Fellow of IEEE. One of his papers won the prestigious 2013 ACM CCS Test-of-Time Award in cybersecurity to recognize its significant impact on the security area over ten years.
Prof. Wai Ho MOW
Title: Bus coding for Low-power High-Speed Interconnects
Abstract: Due to the recent drastic demand on AI accelerator hardware, novel very-large-scale (and even wafer-scale) circuit integration architectures, which may involve 3D stacking mutliple chiplets onto silicon interposers with sophisticated layer interconnections, have been introduced. The capacitive crosstalk of the on-chip bus interconnects induces high power consumption and limits data transmission speed. The classical solution of adding ground shielding is area-inefficient. One of the more area-efficient approaches, called bus coding, is to add one or a few redundant wires which send encoded signals in such a way that the overall latency and/or power consumption is reduced. The most famous single-redundancy-wire bus code is the bus invert code, which has been standardized and adopted in numerous inter-chip bus interconnects applications. In this talk, various known families of low-power bus codes will be surveyed. It will be pointed out that many known bus codes may actually increase, rather than decrease, overall power consumption, after the codec power consumption is taken into consideration. Our recent works on low-power bus codes, which can achieve the state-of-the-art overall power saving, will also be presented.
Bio: Wai Ho Mow received his PhD in Information Engineering from the Chinese University of Hong Kong in 1993. He was an Assistant Professor at the Nanyang Technological University, Singapore, during 1997-1999. He has been with the Hong Kong University of Science and Technology since 2000, and is currently a Professor of ECE. His research areas include communication, coding and information theory. He pioneered the lattice approach (e.g. complex lattice reduction-aided detection) to signal detection problems and gave a unified construction of perfect roots-of-unity (aka CAZAC) sequences which have widespread applications in communication preambles and radar waveforms. He coauthored ~250 journal/conference publications and is a co-inventor of ~40 patents. He served on the editorial boards of seven journals, incl. the IEEE Transactions on Wireless Communications. He is the founding vice chairman and an ex-chairman of the Hong Kong Chapter, IEEE Information Theory Society, and is a past member of the Radio Spectrum Advisory Committee, Office of the Telecommunications Authority of the Hong Kong S.A.R. Government.
Prof. Shao-Lun Huang
Title: A Mathematical Theory to In-Context Learning
Abstract: In-Context Learning (ICL) has emerged as an important new paradigm in natural language processing and large language model (LLM) applications. However, the theoretical understanding of the ICL mechanism remains limited. This talk aims to investigate this issue by studying a particular ICL approach, called concept-based ICL (CB-ICL). In particular, we propose theoretical analyses on applying CB-ICL to ICL tasks, which explains why and when the CB-ICL performs well for predicting query labels in prompts with only a few demonstrations. In addition, the proposed theory quantifies the knowledge that can be leveraged by the LLMs to the prompt tasks, and leads to a similarity measure between the prompt demonstrations and the query input, which provides important insights and guidance for model pre-training and prompt engineering in ICL. Moreover, the impact of the prompt demonstration size and the dimension of the LLM embeddings in ICL are also explored based on the proposed theory. Finally, several real-data experiments are conducted to validate the practical usefulness of CB-ICL and the corresponding theory.
Bio: Shao-Lun Huang received the B.S. degree with honor in 2008 from the Department of Electronic Engineering, National Taiwan University, Taiwan, and the M.S. and Ph.D. degree in 2010 and 2013 from the Department of Electronic Engineering and Computer Sciences, Massachusetts Institute of Technology. From 2013 to 2016, he was working as a postdoctoral researcher jointly in the Department of Electrical Engineering at the National Taiwan University and the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology. Since 2016, he has joined Tsinghua-Berkeley Shenzhen Institute, where he is currently a tenured associate professor. His research interests include information theory, communication theory, machine learning, and statistics.
Prof. Ching-Yi Lai
Title: Accelerated Degeneracy-aware Ordered Statistics Decoding for Quantum Codes via Reliable Subset Reduction
Abstract: Efficient and scalable decoding of quantum codes is essential for quantum error correction.
In this work, we introduce Reliable Subset Reduction (RSR), a reliability-driven preprocessing framework that leverages belief propagation (BP) statistics to identify and remove highly reliable qubits, thereby substantially reducing the effective decoding problem size.
We further identify a degeneracy condition under which high-order ordered statistics decoding (OSD) can be reduced to order-0 OSD.
By integrating this condition with RSR, we develop an Accelerated Degeneracy-aware OSD algorithm (ADOSD).
The resulting BP-ADOSD framework extends naturally to circuit-level noise via detector error models and scales to large instances with more than $10^4$ error variables.
Through extensive simulations, we demonstrate consistently improved decoding performance across a variety of CSS and non-CSS codes under the code-capacity noise model, as well as for rotated surface codes under realistic circuit-level noise.
At low physical error rates, RSR significantly reduces the effective problem size, leading to a substantial reduction in the computational cost of OSD.
These results highlight the practical efficiency and broad applicability of the BP-ADOSD framework for both theoretical and realistic quantum error correction scenarios.
Bio: Ching-Yi Lai received the B.S. and M.S. degrees in Electrical Engineering from National Tsing Hua University, Taiwan, in 2004 and 2006, respectively, and the Ph.D. degree from the University of Southern California in 2013. He is currently an Associate Professor at National Yang Ming Chiao Tung University, Taiwan.
Dr. Lai was the recipient of the Young Scholar Fellowship (Columbus Program) in 2018 and the Excellent Young Research Fellowship in 2021 from the National Science and Technology Council (NSTC), Taiwan. He is currently leading a 5-year NSTC Flagship Quantum Research Project, focusing on the development of fault-tolerant quantum computation architectures. His academic excellence is further recognized by the 2023 IEEE ITSoc & ComSoc Taipei/Tainan Chapter Best Paper Award (Young Scholars) and the 2023 Best Journal Paper Award from the Taiwan Association for Algorithms and Computation Theory. His research interests encompass quantum information theory, quantum coding theory, fault-tolerant quantum computation, and quantum cryptography.
Cover Photo Credit: Kinmen County Government
Page updated
Report abuse