Gal Vardi - A Theory of Learning with Autoregressive Chain of Thought (Heb)
Автор: HUJI Machine Learning Club
Загружено: 2025-11-20
Просмотров: 62
Time and Place
Thursday, November 20th, 2025, 10:30 AM, room C221
Speaker
Gal Vardi (Weizmann)
Title
A Theory of Learning with Autoregressive Chain of Thought
Abstract:
To solve complex tasks, language models produce a Chain-of-Thought leading to the desired answer, where each intermediate token is generated in an autoregressive manner. In this talk, I will present a formal PAC-learning framework for studying this emerging paradigm, both when the chain-of-thought is observed, and when training only on prompt-answer pairs, with the chain-of-thought latent. I will discuss the sample and computational complexity in such settings, and present a simple class of models that allows for efficient universal chain-of-thought learning.
Bio:
Gal is a Senior Scientist in the Department of Computer Science and Applied Mathematics at the Weizmann Institute of Science. Prior to that, he was a postdoctoral researcher at TTI-Chicago, the Hebrew University, and Weizmann. He completed his PhD at the Hebrew University. His research focuses on theoretical machine learning, with an emphasis on deep-learning theory.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
