BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20251106T090819EST-9156HXuv7n@132.216.98.100
DTSTAMP:20251106T140819Z
DESCRIPTION: Title: Towards Efficient and Reliable Generative and Sampling 
 Models\n\nAbstract:\n\nThis talk presents a unified framework for enhancin
 g the reliability and geometric fidelity of generative models. We first de
 velop a diffusion mechanism defined intrinsically on the SE(3) manifold\, 
 enabling the efficient sampling. To address the critical issue of mode col
 lapse in energy-based samplers\, we introduce a novel Importance Weighted 
 Score Matching method that provably improves coverage of complex\, multi-m
 odal distributions. Finally\, we extend these principles to infer underlyi
 ng dynamical systems directly from incomplete and scattered training data.
  Collectively\, this work bridges geometric consistency\, statistical reli
 ability\, and learning from partial observations to advance the frontiers 
 of generative and sampling models.\n\nSpeaker\n\nTianshu Yu is currently a
 n assistant professor at School of Data Science\, The Chinese University o
 f Hong Kong\, Shenzhen (CUHK-Shenzhen). He leads Learning Of Graph & Optim
 ization Lab (LOGO Lab) and is now a visiting assistant professor at MILA -
  Quebec AI Institute. Tianshu serves as the Co-Editor-in-Chief of Science 
 Digitalization (AIMS Publishing). His research interest generally covers M
 achine Learning for Optimization (ML4Opt) and AI4Science\, with a particul
 ar forcus of bringing together machine learning and PDEs.\n\nhttps://mcgil
 l.zoom.us/j/87181846336\n\nMeeting ID: 871 8184 6336\n\nPasscode: None\n
DTSTART:20251107T203000Z
DTEND:20251107T213000Z
LOCATION:Room 1104\, Burnside Hall\, CA\, QC\, Montreal\, H3A 0B9\, 805 rue
  Sherbrooke Ouest
SUMMARY:Tianshu Yu (The Chinese University of Hong Kong\, Shenzhen)
URL:/biology/channels/event/tianshu-yu-chinese-univers
 ity-hong-kong-shenzhen-368694
END:VEVENT
END:VCALENDAR