BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//https://caida.ubc.ca//NONSGML iCalcreator 2.41.92//
CALSCALE:GREGORIAN
METHOD:PUBLISH
UID:65653533-3064-4366-b039-653438376465
X-WR-RELCALID:efc09d74-9c93-479e-a94f-485231ddccde
X-WR-TIMEZONE:America/Vancouver
X-WR-CALNAME:Tuning Free (inference time) Alignment of Large Language Model
 s - Amrit Singh Bedi\, Assistant Professor\, University of Central Florida
BEGIN:VTIMEZONE
TZID:America/Vancouver
TZUNTIL:20261101T090000Z
BEGIN:STANDARD
TZNAME:PST
DTSTART:20241103T020000
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
RDATE:20251102T020000
END:STANDARD
BEGIN:DAYLIGHT
TZNAME:PDT
DTSTART:20240310T020000
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
RDATE:20250309T020000
RDATE:20260308T020000
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:5713ef0e-156d-45c9-abff-23e6e2603953
DTSTAMP:20260416T023541Z
CLASS:PUBLIC
CREATED:20241210T234834Z
DESCRIPTION:Abstract: Traditional fine-tuning of foundation models is compu
 tationally heavy\, involving updates to billions of parameters. A promisin
 g alternative\, alignment via decoding\, adjusts the response distribution
  directly without model updates to maximize a target reward r\, thus provi
 ding a lightweight and adaptable framework for alignment. However\, princi
 pled decoding methods rely on oracle access to an optimal Q-function (Q*)\
 , which is often unavailable in practice. We propose Transfer Q*\, which i
 mplicitly estimates the optimal value function for a target reward through
  a baseline model aligned…
DTSTART;TZID=America/Vancouver:20241216T094500
DTEND;TZID=America/Vancouver:20241216T104500
LAST-MODIFIED:20241210T235227Z
LOCATION:UBC Vancouver Campus\, ICCS X836
SUMMARY:Tuning Free (inference time) Alignment of Large Language Models - A
 mrit Singh Bedi\, Assistant Professor\, University of Central Florida
TRANSP:OPAQUE
URL:https://caida.ubc.ca/index.php/event/tuning-free-inference-time-alignme
 nt-large-language-models-amrit-singh-bedi-assistant
END:VEVENT
END:VCALENDAR