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:20250110T023309Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241203T130000
DTEND;TZID=Asia/Tokyo:20241203T131100
UID:siggraphasia_SIGGRAPH Asia 2024_sess105_papers_854@linklings.com
SUMMARY:MoA: Mixture-of-Attention for Subject-Context Disentanglement in P
 ersonalized Image Generation
DESCRIPTION:Technical Papers\n\nKuan-Chieh Wang, Daniil Ostashev, Yuwei Fa
 ng, Sergey Tulyakov, and Kfir Aberman (Snap Inc.)\n\nWe introduce a new ar
 chitecture for personalization of text-to-image diffusion models, coined M
 ixture-of-Attention (MoA). Inspired by the Mixture-of-Experts mechanism ut
 ilized in large language models (LLMs), MoA distributes the generation wor
 kload between two attention pathways: a personalized branch and a non-pers
 onalized prior branch.\nMoA is designed to retain the original model's pri
 or by fixing its attention layers in the prior branch, while minimally int
 ervening in the generation process with the personalized branch that learn
 s to embed subjects in the layout and context generated by the prior branc
 h.\nA novel routing mechanism manages the distribution of pixels in each l
 ayer across these branches to optimize the blend of personalized and gener
 ic content creation. \nOnce trained, MoA facilitates the creation of high-
 quality, personalized images featuring multiple subjects with compositions
  and interactions as diverse as those generated by the original model.\nCr
 ucially, MoA enhances the distinction between the model's pre-existing cap
 ability and the newly augmented personalized intervention, thereby offerin
 g a more disentangled subject-context control that was previously unattain
 able.\n\nRegistration Category: Full Access, Full Access Supporter\n\nLang
 uage Format: English Language\n\nSession Chair: Kfir Aberman (Snap)
URL:https://asia.siggraph.org/2024/program/?id=papers_854&sess=sess105
END:VEVENT
END:VCALENDAR