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-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:20251223T143000
DTEND;TZID=Asia/Hong_Kong:20251223T153000
DTSTAMP:20260511T134233
CREATED:20251217T024739Z
LAST-MODIFIED:20251217T024739Z
UID:114432-1766500200-1766503800@ece.hku.hk
SUMMARY:RPG Seminar – Design of High-Performance and Energy-Efficient AI Accelerator for Edge Computing
DESCRIPTION:Abstract\nThis seminar provides a systematic showcase of our end-to-end energy-efficient solutions in AI accelerator design\, spanning from the foundational compute unit to specialized architectures and large model deployment on edge devices. We begin by introducing a novel Reconfigurable Processing Element (PE)\, the core compute unit designed to support multiple floating-point (e.g.\, BF16\, FP16) and fixed-point (e.g.\, INT8\, INT4) Multiply-Accumulate (MAC) operations with 100% hardware utilization\, achieving exceptional energy efficiency exceeding 1700 GFLOPS/W for deep learning tasks. Subsequently\, addressing the need for extreme efficiency in edge vision\, we present dedicated\, multiplication-free Look-Up Table (LUT) accelerators like BDLUT (for blind denoising)\, EdgeLUT (for all fixed-resolution image restoration) and ScaleLUT (for real-time 4K super-resolution). These designs replace traditional convolutional operations with efficient LUT inference\, resulting in significantly lower hardware and power consumption. Finally\, we focus on high-efficiency acceleration for complex modern AI models: we propose the QuadINR framework\, which utilizes hardware-efficient piecewise quadratic activation functions for Implicit Neural Representations (INR). For Large Language Models (LLMs)\, EdgeLLM is a highly efficient CPU-FPGA heterogeneous accelerator that employs a mixed-precision PE array (FP16/FP16 and FP16/INT4) and a unified data parallelism scheme\, successfully tackling LLM deployment challenges to achieve up to 1.91x higher throughput and 7.55x better energy efficiency than commercial GPUs. In summary\, this presentation offers a comprehensive view of our complete technological roadmap in the world of efficient AI chips\, achieved through co-optimization of hardware and algorithms\, to realize high-performance and energy-efficient AI acceleration for edge computing. \nSpeaker\nMr. Boyu Li\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nBiography of the Speaker\nBoyu Li is a Ph.D. candidate in the Department of Electrical and Electronic Engineering at The University of Hong Kong\, under the supervision of Professor Ngai Wong. His research focuses on deep learning\, reconfigurable computing\, and the design of AI accelerators. \nOrganiser\nProf. Ngai Wong\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20251223/
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
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20251223T143000
DTEND;TZID=Asia/Hong_Kong:20251223T153000
DTSTAMP:20260511T134233
CREATED:20251219T110207Z
LAST-MODIFIED:20251219T110610Z
UID:114486-1766500200-1766503800@ece.hku.hk
SUMMARY:RPG Seminar – Design of Novel Structured Light Beams in Optical Manipulation and Two-Photon Microscopy
DESCRIPTION:Zoom Link:\nhttps://hku.zoom.us/j/6983793721?pwd=UjwaYARlhjvhUf7DDp3bMjbJnlhVbr.1 \nAbstract\nStructured illumination refers to a class of optical imaging methods in which the sample is illuminated with a controlled spatial light pattern rather than uniform illumination. By altering the illumination pattern\, it can facilitate the extraction of spatial or temporal information from the sample\, and may improve imaging performance such as resolution or acquisition speed. It also enables precise manipulation of microscopic objects through targeted light patterns. One example is the higher-order BBs generated by adding a vortex phase to the zero-order BBs and exhibiting a ring-shaped intensity distribution. Their characteristic length results in different trapping states of particles with different sizes\, which is of assistance to the size-sorting process. To this date\, there has been no research on high-order conveyor beams with a vortex modulation. In this study\, a spatial light modulator (SLM) is used to generate the ring-shaped conveyor beams. The beam profile is simulated and then confirmed with a two-photon microscopy system. The conveyor beam shows two alternating rings with an axial period. This technique opens a door to simultaneous optical manipulation of the particle’s OAM and longitudinal translation. The application of another class of beams\, axially encoded beams\, in multiphoton microscopy (MPM)\, will also be introduced. Combined with computational decoding\, these beams enable axially parallel excitation (APEX) that extends per-plane exposure without slowing acquisition or requiring higher power. High-speed volumetric MPM with improved contrast will be demonstrated. \nSpeaker\nMs. Minghui SHI\nDepartment of Electrical and Electronic Engineering\,\nThe University of Hong Kong \nSpeaker’s Biography\nMinghui SHI received the B.S. degree in optical information science and technology from the Beijing University of Technology\, Beijing\, China\, in 2021. She is currently working toward the Ph.D. degree with the Department of Electrical and Electronic Engineering\, The University of Hong Kong\, Hong Kong. Her research interests include structured illumination\, fluorescence imaging\, and mid-infrared laser applications. \nOrganiser\nProf. Kenneth K.Y. WONG\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong
URL:https://ece.hku.hk/events/20251223-2/
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