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:20240101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20251210T150000
DTEND;TZID=Asia/Hong_Kong:20251210T160000
DTSTAMP:20260511T134219
CREATED:20251202T025642Z
LAST-MODIFIED:20251202T025642Z
UID:114328-1765378800-1765382400@ece.hku.hk
SUMMARY:RPG Seminar – Mamba model acceleration on RRAM-Based Compute-in-Memory (CIM) Systems integrated with Selective State-Space Streaming
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/97793742616?pwd=YIyYlokhzOsap3IvbsbwmfaHVHFoin.1 \nAbstract\nAs Generative AI shifts toward handling massive context windows\, the quadratic complexity of Transformer architecture has become a significant bottleneck. State Space Models (SSMs)\, particularly Mamba\, have emerged as a promising solution\, offering linear-time scaling and superior efficiency. However\, the unique computational duality of SSMs—requiring both memory-intensive projections and agile\, input-dependent state updates—presents new challenges that traditional von Neumann architectures and GPUs struggle to address efficiently. \nThis seminar explores the evolution of efficient sequence modeling and the critical hardware innovations required to support it. We will examine the “Memory Wall” problem in modern AI deployment and introduce Compute-in-Memory (CIM) using Resistive RAM (RRAM) as a paradigm shift to minimize data movement. The discussion will focus on the principles of hardware-software co-design\, illustrating how tailored architecture can bridge the gap between memory-bound operations and dynamic recursions. By integrating specialized streaming dataflows with non-volatile memory technologies\, we can define a new computational fabric capable of enabling the next generation of energy-efficient edge AI. \nSpeaker\nMr. Mingzi Li\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nBiography of the Speaker\nMr. Mingzi Li is currently pursuing the Ph.D. degree with the Department of Electrical and Electronic Engineering\, supervised by Professor Han Wang. He received his B.Eng. in Computer Engineering from The Chinese University of Hong Kong in 2021 and the M.S. in Electrical and Electronic Engineering from The University of Hong Kong in 2022. His research interests include compute-in-memory architectures\, RRAM-based systems\, hardware acceleration for emerging sequence models and efficient AI systems. \nOrganiser\nProf. Han Wang\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20251210/
LOCATION:Online via Zoom
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
END:VCALENDAR