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:20251201T103000
DTEND;TZID=Asia/Hong_Kong:20251201T113000
DTSTAMP:20260511T134220
CREATED:20251125T033044Z
LAST-MODIFIED:20251125T033044Z
UID:114263-1764585000-1764588600@ece.hku.hk
SUMMARY:RPG Seminar – High-throughput Neuromorphic Computational Imaging
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/99712347936?pwd=P2oHpewBKizDaTJNY9m4YowNQLZfaP.1 \nAbstract\nHigh-throughput dynamic imaging must recover fine spatial structure under rapid motion\, yet no conventional sensor can fully overcome the trade-offs between spatial resolution\, temporal resolution\, and motion-induced degradation. Frame sensors inevitably blur fast dynamics due to global integration\, while event sensors\, although extremely fast and high-dynamic-range\, provide only local 1-bit temporal changes and lack global spatial context. These sensing limitations fundamentally constrain applications ranging from defect inspection to phase-flow analysis. In this seminar\, I will present a neuromorphic computational imaging paradigm\, Neuromorphic Super-Resolution (NeuroSR)\, that addresses these limitations through physics-informed spatio-temporal feature inference. NeuroSR unifies the complementary measurements of frames and events into a fully differentiable architecture\, enabling high space–time resolved reconstruction and direct inference of physical structure such as motion blur kernels or coherent wave propagation. To illustrate the generality of this paradigm\, I will also introduce Neuromorphic Wave-Normal Sensing (NeuroSH) as a representative white-box example. NeuroSH demonstrates how asynchronous event cues can recover large-gradient wavefront information and surpass classical spot-overlapping constraints in dynamic wavefront sensing systems. Together\, these results highlight a unified neuromorphic approach that transforms both dynamic imaging and physical-structure inference\, enabling ultrafast defect inspection\, large-gradient wavefront analysis\, and high-throughput computational imaging well beyond the limits of conventional sensors. \nSpeaker\nMr. Chutian Wang\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nBiography of the Speaker\nChutian Wang received the B.S. degree from the University of Science & Technology Beijing in 2020\, and the M.S. degree at Imperial College London in 2021. He is currently working towards his Ph.D. degree with the Department of Electrical and Electronic Engineering\, the University of Hong Kong. His research interests include computational neuromorphic imaging\, wavefront sensing and digital holography. \nOrganiser\nProf. Edmund Y. Lam\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20251201-3/
LOCATION:Online via Zoom
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:20251201T140000
DTEND;TZID=Asia/Hong_Kong:20251201T150000
DTSTAMP:20260511T134220
CREATED:20251119T033808Z
LAST-MODIFIED:20251119T033808Z
UID:113994-1764597600-1764601200@ece.hku.hk
SUMMARY:RPG Seminar – Brain-inspired Random Memristors Pruning for Input-aware Dynamic SNN
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/96497087839?pwd=5X1msaxhZNiH87SuzGTPgQHZILJmgi.1 \nAbstract\nMachine learning has advanced unprecedentedly\, exemplified by GPT-4 and SORA. However\, they cannot parallel human brains in efficiency and adaptability due to differences in signal representation\, optimization\, run-time reconfigurability\, and hardware architecture. To address these challenges\, we introduce PRIME—a pruning optimization for input-aware dynamic memristive spiking neural networks. PRIME leverages spiking neurons to emulate biological spiking mechanisms and optimizes the topology of random memristive SNNs\, mitigating memristor programming stochasticity. Additionally\, it employs an input-aware early-stop policy to reduce latency and memristive in-memory computing to alleviate the von Neumann bottleneck. Validated on a memristor-based macro\, PRIME achieves competitive classification accuracy and superior energy efficiency. \nSpeaker\nMr. Bo Wang\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nBiography of the Speaker\nBo Wang received B.Eng. degree in Power Engineering\, Beihang University\, Beijing\, China\, in 2020\, and M.Eng. degree in Pattern Recognition and Intelligent Systems\, Beihang University\, Beijing\, China\, in 2022. He is currently pursuing the Ph.D. degree with the Department of Electrical and Electronic Engineering under the supervision of Prof. Xiaojuan Qi. His research interests mainly include in-memory computing\, Embodied AI and software-hardware co-design. \nOrganiser\nProf. Xiaojuan Qi\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20251201/
LOCATION:Online via Zoom
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:20251201T150000
DTEND;TZID=Asia/Hong_Kong:20251201T160000
DTSTAMP:20260511T134220
CREATED:20251124T035753Z
LAST-MODIFIED:20251124T035753Z
UID:114201-1764601200-1764604800@ece.hku.hk
SUMMARY:RPG Seminar – Efficient Learning for Image Restoration and Single-Photon Imaging without Clean Data
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/92646013468?pwd=lkoH511LkjLHtW43awHeBpEVnLfZ7b.1 \nAbstract\nSupervised deep learning has revolutionized computational imaging but relies heavily on vast datasets of clean\, ground-truth images\, which are often challenging to acquire in practice. This seminar presents a series of methods that break this dependency by embracing weakly-supervised and unsupervised learning\, directly addressing the challenge of learning without clean data. First\, I will introduce a Fourier-based statistical equivalence between learning with noisy targets and clean targets. Building on this\, I will present a weakly supervised framework for diverse image restoration tasks\, along with two unsupervised denoising methods specifically designed for pixel-wise and stripe-wise noise. Finally\, I will introduce a physics-informed unsupervised framework that can enable image restoration learning for single photon imaging with only the training data degraded by the blurring effect\, Poisson noise\, and readout noise. Collectively\, this seminar demonstrates powerful and flexible learning paradigms that advance the computational imaging for scenarios where clean data is unavailable. \nSpeaker\nMr. Haosen Liu\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nBiography of the Speaker\nHaosen Liu received his B.Sc. and M.S. degrees from Huazhong University of Science and Technology\, and is currently a fourth-year Ph.D. candidate in the Department of Electrical and Electronic Engineering at The University of Hong Kong under the supervision of Prof. Edmund Y. Lam. His research interests mainly include data-efficient deep learning methods for image restoration and computational imaging. \nOrganiser\nProf. Edmund Y. Lam\nDepartment of Electrical and Electronic Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20251201-2/
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