BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Understanding and Improving Reasoning in Large Lang
 uage Models
DTSTART:20260226T101500
DTEND:20260226T111500
DTSTAMP:20260916T080859Z
UID:1abdf7615f980c9b7d15e21e3ba33c664fa3203bce2b3bb1145af960
CATEGORIES:Conferences - Seminars
DESCRIPTION:Par : Nouha Dziri - Allen Institute for AI\nIC Faculty candida
 te\n\nAbstract\nDespite the impressive capabilities of AI models\, they pr
 esent a striking paradox: they can solve Olympiad-level math problems whil
 e still making basic reasoning errors that even non-experts would avoid. T
 his points to fundamental limitations in their ability to generalize.\nIn 
 this talk\, I will share key insights into the strengths and limitations o
 f large language models\, examine when reinforcement learning improves rea
 soning and when it struggles to generalize\, and explore approaches to enh
 ance their reasoning capabilities. I will also argue that many safety fail
 ures stem from underlying reasoning failures\, and discuss how to make mod
 els more robust to adversarial attacks. I will conclude with a forward-loo
 king research agenda: scaling reasoning through smarter algorithms and tra
 ining recipes\, rigorous evaluation frameworks to measure real progress\, 
 and ensuring safety at every stage as models become increasingly autonomou
 s.\n\nBio\nNouha Dziri is a Senior Research Scientist at Ai2. Her research
  spans a wide range of problems in AI\, with a focus on building and impro
 ving large language models (LLMs).  She co-led the post-training effort f
 or the OLMo models. Prior to that\, she was a postdoctoral researcher work
 ing with Yejin Choi at Ai2. She has also interned at Google DeepMind\, Mic
 rosoft Research\, and Mila. She is the recipient of multiple awards\, incl
 uding Best Paper Awards at NeurIPS 2025 and NAACL 2025. Her work has been 
 featured in leading media outlets including The Economist\, TechCrunch\, L
 e Monde\, Science\, and Quanta Magazine. She has delivered invited talks a
 t top universities such as Stanford\, Oxford\, Cambridge\, Edinburgh\, Pri
 nceton\, McGill\, and CMU. She has also been a keynote speaker and a panel
 ist at workshops at major AI conferences including NeurIPS\, ICLR\, and IC
 ML. She earned her PhD from the University of Alberta and the Alberta Mach
 ine Intelligence Institute (Amii) in 2023.\n\nMore information\n\n 
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/65062323908
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR