BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系 - ECPv6.15.20//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系
X-ORIGINAL-URL:https://ece.hku.hk
X-WR-CALDESC:Events for Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:20250101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20260513T100000
DTEND;TZID=Asia/Hong_Kong:20260513T110000
DTSTAMP:20260510T221559
CREATED:20260507T082216Z
LAST-MODIFIED:20260507T082216Z
UID:115875-1778666400-1778670000@ece.hku.hk
SUMMARY:RPG Seminar – Towards Controllable and Cinematic Visual Content Generation
DESCRIPTION:Zoom Link \nhttps://hku.zoom.us/j/96029601263 \nAbstract\nThis seminar presents novel solutions to bridge the gap between creative intent and automated visual synthesis. We first address the challenge of compositional alignment in text-to-image generation. This includes T2I-CompBench(++) for systematic evaluation of multi-attribute prompts\, and GenMAC\, a multi-agent collaborative framework that enables precise semantic orchestration. Second\, we target the lack of cinematic artistry in video generation via Filmaster\, a system that injects professional camera language and rhythm into the synthesis process using real-film references. Finally\, we tackle spatio-temporal inconsistency in dynamic sequences using CineScene\, which leverages implicit 3D-aware representations to preserve scene consistency across camera trajectories. Together\, these contributions pave the way for next-generation controllable tools in digital cinematography and professional content creation. \nSpeaker\nMs Kaiyi HUANG\nDepartment of Electrical and Computer Engineering\nThe University of Hong Kong \nBiography of the Speaker\nKaiyi Huang received her BSc degree and MSc in Electrical Engineering from the Shanghai Jiao Tong University\, and a double MSc degree from Ecole Centrale Paris (joint program). She is currently pursuing a PhD in the Department of Electrical and Computer Engineering at the University of Hong Kong. Her research focuses on the image/video generation\, film generation and world model. \nOrganiser\nProf Xihui LIU \nDepartment of Electrical and Computer Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20260513/
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20260514T103000
DTEND;TZID=Asia/Hong_Kong:20260514T113000
DTSTAMP:20260510T221559
CREATED:20260508T020416Z
LAST-MODIFIED:20260508T020416Z
UID:115879-1778754600-1778758200@ece.hku.hk
SUMMARY:RPG Seminar – HIMSA: A Heterogeneous In-Memory Computing and Searching Architecture for Efficient Attention-Based Models
DESCRIPTION:Zoom Link \nhttps://hku.zoom.us/j/99174148480?pwd=duVxaYZOJDT6MWxDh4OKOMmyo12A7A.1 \nAbstract\nThe Transformer architecture\, the foundation for modern large language models (LLMs)\, has revolutionized natural language processing and other AI domains. However\, its significant computational and memory requirements\, primarily from the matrix multiplication in the self-attention mechanism\, present major challenges for conventional hardware. While intensive research on in-memory computing (IMC) technology offers a path to overcome the memory bottleneck\, using IMC for Transformers remains challenging. This is because the dynamic matrix multiplication with frequently changing Key\, Query\, and Value matrices require frequent and costly write operations that are ill-suited for non-volatile memories (NVM) technologies like ReRAM. This work introduces HIMSA Heterogeneous In-Memory Computing and Searching Architecture\, which employs vector quantization on K and V matrices. This technique transforms the dynamic vector-matrix multiplications into static operations performed on pre-trained codebooks\, thereby eliminating the need for runtime write operations in the attention mechanism. The proposed architecture was evaluated through circuit-level simulations that account for the peripheral circuit designs\, including nearest neighbor search and the division-less Softmax operations. More importantly\, its write-free attention mechanism mitigates the concerns over limited write endurance of ReRAM devices . This work presents a promising pathway toward highly efficient NVM-based hardware acceleration for next-generation AI models. \nSpeaker\nMr Muyuan PENG\nDepartment of Electrical and Computer Engineering\nThe University of Hong Kong \nBiography of the Speaker\nPeng Muyuan received the B.S. degree at the University of Science and Technology of China majored in Applied Physics. He is currently pursuing the Ph.D. degree in electrical and electronic engineering at the Department of Electrical and Electronic Engineering\, The University of Hong Kong. His current research interests include non-volatile memory devices\, in-memory computing and related neural network accelerations. \nOrganiser\nProfessor Can LI \nDepartment of Electrical and Computer Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20260514-2/
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20260514T133000
DTEND;TZID=Asia/Hong_Kong:20260514T143000
DTSTAMP:20260510T221559
CREATED:20260505T021413Z
LAST-MODIFIED:20260505T021413Z
UID:115830-1778765400-1778769000@ece.hku.hk
SUMMARY:RPG Seminar – Cooperative Edge AI: From Event-triggered Inference to Efficient Model Downloading
DESCRIPTION:Zoom Link \nhttp://hku.zoom.us/j/7074144117?omn=95813783034 \nAbstract\nCooperative edge AI enables edge devices and edge servers to collaboratively execute intelligent tasks under limited computation\, storage\, energy\, and communication resources. In this talk\, we discuss two complementary research directions toward communication-efficient cooperative edge AI. First\, we introduce an event-triggered cooperative inference framework for rare-event detection in edge intelligence systems. Rare events are usually infrequent but highly critical\, while conventional edge inference systems may overlook them due to data imbalance and rigid resource allocation. To address this issue\, a dual-threshold multi-exit architecture is adopted\, allowing confident normal events to be processed locally while complex or uncertain rare events are selectively offloaded to the edge server for more accurate classification. Second\, we present an efficient AI model downloading framework based on parametric-sensitivity-aware retransmission. Instead of treating all model parameters equally\, this framework exploits the unequal importance of neural network parameters and allocates wireless retransmission resources to more sensitive model packets. In this way\, downloading latency can be reduced while inference performance is preserved. The talk concludes with a discussion of future research directions in cooperative edge AI\, highlighting open challenges and opportunities in communication-efficient inference\, adaptive model deployment\, and resource-aware edge intelligence. \nSpeaker\nMr Zhou You\nDepartment of Electrical and Computer Engineering\nThe University of Hong Kong \nBiography of the Speaker\nZhou You is currently pursuing a Ph.D. degree in the Department of Electrical and Electronic Engineering at The University of Hong Kong\, under the supervision of Prof. Kaibin Huang. He received his B.Eng. degree in Electrical Engineering from the University of Wisconsin–Madison\, USA\, in 2021. His research interests include wireless communications\, edge inference\, and AI model downloading. \nOrganiser\nProf. Kaibin HUANG \nDepartment of Electrical and Computer Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20260514/
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20260519T160000
DTEND;TZID=Asia/Hong_Kong:20260519T170000
DTSTAMP:20260510T221559
CREATED:20260420T064633Z
LAST-MODIFIED:20260420T064633Z
UID:115720-1779206400-1779210000@ece.hku.hk
SUMMARY:RPG Seminar – From Understanding to Intervention: Interpretability-Guided Methods for Improving Large Language Models
DESCRIPTION:  \nAbstract\nLarge language models have achieved impressive performance\, but improving them efficiently and reliably requires more than scaling alone. In this talk\, I present a series of works that explore how internal understanding of LLMs can be translated into practical interventions for better capability\, efficiency\, and controllability. I begin with actionable mechanistic interpretability\, introducing a unified “Locate\, Steer\, and Improve” perspective that turns model analysis into a framework for intervention. I then show how this perspective supports several concrete advances: data-free mixed-precision quantization guided by numerical and structural sensitivity\, multilingual capability enhancement through representation shifting and contrastive alignment\, personalized multi-teacher distillation that routes each prompt to its most suitable teacher\, and coarse-to-fine selective fine-tuning for mitigating catastrophic forgetting while preserving general versatility. Together\, these works reflect a common theme: interpretability is not only a tool for explaining LLMs\, but also a principled basis for designing more efficient training\, compression\, and adaptation methods. \nSpeaker\nMr Hengyuan ZHANG\nDepartment of Electrical and Computer Engineering\nThe University of Hong Kong \nBiography of the Speaker\nHengyuan Zhang is a Ph.D. candidate at the University of Hong Kong\, supervised by Prof. Ngai Wong and Prof. Hayden Kwok-Hay So. His research focuses on the improvement of efficiency and interpretability within large language models. He aims to uncover and characterize the internal processes that govern model behavior\, with the goal of improving model speciality\, interpretability\, and reliability in real-world deployments. He has published multiple papers in leading venues such as ACL\, EMNLP\, TKDD\, and NeurIPS. \nOrganiser\nProf. Ngai WONG \nDepartment of Electrical and Computer Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20260519/
LOCATION:Room CB-603\, 6/F\, Chow Yei Ching Building\, The University of Hong Kong
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
END:VCALENDAR