BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Two New Insights into Beam Search
DTSTART:20211011T161500
DTEND:20211011T171500
DTSTAMP:20260925T054658Z
UID:59115422d7cdd77bfbf9b7f168c2bbfafda047915639cb9ffe0128df
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Ryan Cotterell - ETH Zurich\nVideo of his talk\n\nAbstract
 \nAs a simple search heuristic\, beam search has been used to decode model
 s developed by the NLP community for decades. Indeed\, it is noteworthy th
 at beam search is one of the few NLP algorithms that has stood the test of
  time: It has remained a cornerstone of NLP systems since the 1970s (Reddy
 \, 1977). As such\, beam search became the natural choice for decoding neu
 ral probabilistic text generators—whose design makes evaluating the full
  search space impossible While there is no formal guarantee that beam sear
 ch will return—or even approximate—the highest-scoring candidate under
  a model\, it has repeatedly proven its merit in practice and\, thus\, has
  largely been tolerated—even embraced—as NLP’s go-to search heuristi
 c. This talk further embraces beam search. We discuss two novel formal ins
 ights into beam search. In the first act\, we discuss an algorithmic advan
 ce that allows beam search to be prioritized\, i.e. it returns the best hy
 pothesis (modulo the beam size) first. Our algorithmic extension yields a 
 Dijkstra-ified beam search that provably emulates standard beam search. In
  the second act\, we draw a connection between the uniform information den
 sity hypothesis from cognitive science and beam search’s efficacy as a s
 earch heuristic. We offer a linguistic reason why beam search may work so 
 well in practice even though\, as an approximation to the argmax\, it may 
 be arbitrarily bad. The work described in this talk is described in public
 ations at TACL (20200 and EMNLP (2020) and won an honorable mention for be
 st paper at the latter.\n\nBio\nRyan Cotterell completed his undergraduate
  degree at Johns Hopkins University in Cognitive Science (with focal areas
  in Linguistics and Computational Methods) under the tutelage of Colin Wil
 son. He was then recruited by Jason Eisner to do a Ph.D. in Computer Scien
 ce at Johns Hopkins University where he was a member of the Center for Lan
 guage and Speech Processing. He is currently a tenure-track assistant prof
 essor at ETH Zürich in the Department of Computer Science where he is a m
 ember of the Institut für maschinelles Lernen. He was previously a Lectur
 er at the Computer Laboratory at the University of Cambridge in the United
  Kingdom where he is still affiliated. He has also done research stints at
  Google AI\, Facebook AI Research and the Ludwig Maximilian University of 
 Munich.\n\nMore information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/65309270257?pwd=REhaQW9QRVhtbTN5akZpVHRwZHlOQT09
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR