BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:AI Center Seminar - AI Fundamentals series - Dr. Tiago Pimentel
DTSTART:20260224T100000
DTEND:20260224T120000
DTSTAMP:20260415T080118Z
UID:5dd8a96b2d6000620f8a92b94afbeb0befcff27e26f6f60af3eb8216
CATEGORIES:Conferences - Seminars
DESCRIPTION:Tiago Pimentel\nThe talk is organized by the EPFL AI Center
  as part of the AI fundamentals seminar series.\n\nHosting professor: Pro
 f. Robert West (DLAB)\n\nTitle\nPromises and Limitations of Causality for 
 Machine Learning Interpretability\n\nAbstract\nHow can we move from observ
 ing what a model does to understanding why it does it? In this talk\, I ar
 gue that causality is the key to uncovering the mechanisms underlying mode
 l predictions. First\, I examine a “macro” view of model analysis\, sh
 owing how econometric tools—such as regression discontinuity or differen
 ce-in-differences—can isolate the causal impact of specific design choic
 es\, like tokeniser and training data selection\, on a model’s outputs. 
 Second\, I turn to a “micro” view of mechanistic interpretability\, fo
 cusing on causal abstraction as a method to verify if a model implements a
  high-level algorithm. I demonstrate that this approach faces a critical l
 imitation: without strict assumptions about how models encode information\
 , the framework becomes vacuous\, implying that any model implements any a
 lgorithm. This reveals that the ability to predictably intervene on a mode
 l is not\, on its own\, sufficient to guarantee we understand it. I conclu
 de with a short discussion about how causality can be used to develop more
  principled interpretability methods.\n\nBio\nTiago Pimentel is a Postdoct
 oral Researcher at ETH Zürich\, working in machine learning interpretabil
 ity and psycholinguistics. His long-term goal is to understand how humans 
 and machines process language. To this end\, his research adopts an interd
 isciplinary approach\, leveraging information theory and causality to stud
 y the mechanisms behind model behaviour and human cognition.\n\n 
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117 https://epfl.zoom.u
 s/j/61181586850?pwd=e2eowYHb3WBIQWqCiSKUaK4IiRAKBs.1
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR