BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Just asking questions (MIT) - Jacob Andreas (MIT)
DTSTART:20251120T140000Z
DTEND:20251120T150000Z
UID:TALK237778@talks.cam.ac.uk
CONTACT:Lucas Resck
DESCRIPTION:Abstract:\nIn the age of deep networks\, "learning" almost inv
 ariably means "learning from examples". We train language models with huma
 n-generated text and labeled preference pairs\, image classifiers with lar
 ge datasets of images\, and robot policies with rollouts or demonstrations
 . When human learners acquire new concepts and skills\, we often do so wit
 h richer supervision\, especially in the form of language---we learn new c
 oncepts from examples accompanied by descriptions or definitions\, and new
  skills from demonstrations accompanied by instructions. Crucially\, langu
 age-based supervision involves not only instructions but *questions*---stu
 dents ask questions to elicit the most useful pieces of supervision\, and 
 teachers ask questions to probe student knowledge and encourage them to ac
 quire new skills or aspects of understanding on their own. This talk will 
 focus on a few recent projects focused on building computational models th
 at can ask good questions for both learning and teaching\, with applicatio
 ns spanning LM alignment\, policy learning\, and education. This is joint 
 work with Belinda Li\, Alex Tamkin\, Noah Goodman\, Andi Peng\, Ilia Sucho
 lutsky\, Nishanth Kumar\, Julie A Shah\, Andreea Bobu\, Alexis Ross\, Gabe
  Grand\, Valerio Pepe and Josh Tenenbaum.\n\nBio: Jacob Andreas is an asso
 ciate professor at MIT in EECS and CSAIL. He completed his PhD at Berkeley
 \, where he was a member of the Berkeley NLP Group and the Berkeley AI Res
 earch Lab. He has also spent time with the Cambridge NLIP Group\, the NLP 
 Group\, and the former Center for Computational Learning Systems at Columb
 ia. His research focuses on language as a communicative and computational 
 tool. He studies how people learn to understand and generate novel utteran
 ces with remarkably little data and how language facilitates the acquisiti
 on of new concepts and reasoning. Recognizing that current machine learnin
 g techniques fall short of human abilities in language understanding and l
 earning from language\, his work aims to uncover the computational foundat
 ions of language learning and develop general-purpose intelligent systems 
 capable of effective human communication and learning from human guidance.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR