BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Decoding Strategies for Large Language Models
DTSTART:20230810T140000
DTEND:20230810T160000
DTSTAMP:20260407T101350Z
UID:b2ad6fa03f0b6b665ceee148cd7fae0c0518988902e17194db8020f0
CATEGORIES:Conferences - Seminars
DESCRIPTION:Saibo Geng\nEDIC candidacy exam\nExam president: Prof. Antoine
  Bosselut\nThesis advisor: Prof. Robert West\nCo-examiner: Prof. Viktor Ku
 ncak\n\nAbstract\nLarge Language Models (LLMs) have significantly advanced
  the field of artificial intelligence\, achieving state-of-the- art result
 s across a diverse range of tasks. Central to their success is the decodin
 g algorithm\, which converts the model’s probability distribution into t
 he generated output text. This report first reviews three notable decoding
  algorithms: constrained decoding\, diverse beam search\, and a recent app
 roach for handling multi- step reasoning tasks with LLMs. We then propose 
 a novel area of investigation—designing a decoding algorithm specificall
 y to enhance the multi-step reasoning capabilities of LLMs.\n\nBackground 
 papers\n- GenIE: Generative Information Extraction: https://arxiv.org/ab
 s/2112.08340\n- Diverse Beam Search: Decoding Diverse Solutions from Neur
 al Sequence Models: https://arxiv.org/abs/1610.02424\n- Tree of Thoughts
 : Deliberate Problem Solving with Large Language Models: https://arxiv.or
 g/abs/2305.10601\n\n\n 
LOCATION:BC 333 https://plan.epfl.ch/?room==BC%20333
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR