BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260805T205600Z
LOCATION:Room 403 A
DTSTART;TZID=America/Los_Angeles:20260721T144000
DTEND;TZID=America/Los_Angeles:20260721T145000
UID:siggraph_SIGGRAPH 2026_sess104_papers_482@linklings.com
SUMMARY:MultiAct: Text-to-Motion Generation from Composite Text via Tailor
 ed Attention Guidance
DESCRIPTION:Nathan Sala (Tel Aviv University); Ofir Abramovich and Ariel S
 hamir (Reichman University); Daniel Cohen-Or (Tel Aviv University); Andrea
 s Aristidou (University of Cyprus, CYENS Centre of Excellence); and Sigal 
 Raab (Tel Aviv University)\n\nText-to-motion models struggle with composit
 e prompts, often collapsing to a single action. We present MultiAct, an in
 ference-time, unpaired framework that enhances pretrained generators by am
 plifying cross-attention for underrepresented tokens. A lightweight decisi
 on scheme selects optimal parameters per prompt, improving semantic covera
 ge while preserving motion realism without retraining or architectural cha
 nges.\n\nInterest Area: Research & Education\n\nKeyword: Animation, Artifi
 cial Intelligence/Machine Learning, Robotics, Simulation\n\nRegistration C
 ategory: Full Conference Supporter, Full Conference\n\nSession Chair: Yuti
 ng Ye (Reality Labs Research, Meta; Meta)\n\n
END:VEVENT
END:VCALENDAR
