BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171537Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241206T093400
DTEND;TZID=Asia/Tokyo:20241206T094600
UID:siggraphasia_SIGGRAPH Asia 2024_sess141_papers_776@linklings.com
SUMMARY:Accelerate Neural Subspace-Based Reduced-Order Solver of Deformabl
 e Simulation by Lipschitz Optimization
DESCRIPTION:Aoran Lyu (South China University of Technology, University of
  Manchester); Shixian Zhao, Chuhua Xian, Zhihao Cen, and Hongmin Cai (Sout
 h China University of Technology); and Guoxin Fang (Chinese University of 
 Hong Kong)\n\nReduced-order simulation is an emerging method for accelerat
 ing physical simulations with high DOFs, and recently developed neural-net
 work-based methods with nonlinear subspaces have been proven effective in 
 diverse applications as more concise subspaces can be detected. However, t
 he complexity and landscape of simulation objectives within the subspace h
 ave not been optimized, which leaves room for enhancement of the convergen
 ce speed. This work focuses on this point by proposing a general method fo
 r finding optimized subspace mappings, enabling further acceleration of ne
 ural reduced-order simulations while capturing comprehensive representatio
 ns of the configuration manifolds. We achieve this by optimizing the Lipsc
 hitz energy of the elasticity term in the simulation objective, and incorp
 orating the cubature approximation into the training process to manage the
  high memory and time demands associated with optimizing the newly introdu
 ced energy. Our method is versatile and applicable to both supervised and 
 unsupervised settings for optimizing the parameterizations of the configur
 ation manifolds. We demonstrate the effectiveness of our approach through 
 general cases in both quasi-static and dynamics simulations. Our method ac
 hieves acceleration factors of up to 6.83 while consistently preserving co
 mparable simulation accuracy in various cases, including large twisting, b
 ending, and rotational deformations with collision handling. This novel ap
 proach offers significant potential for accelerating physical simulations,
  and can be a good add-on to existing neural-network-based solutions in mo
 deling complex deformable objects.\n\nRegistration Category: Full Access, 
 Full Access Supporter\n\nLanguage Format: English Language\n\nSession Chai
 r: Sheldon Andrews (École de Technologie Supérieure (ÉTS), McGill Universi
 ty)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_776&sess=sess141
END:VEVENT
END:VCALENDAR