/Kolo-Materialy

Primary LanguageJupyter Notebook

Materiały Koła Naukowego Uczenia Maszynowego UW

Materiały ze spotkań (lista niekompletna)

2018/2019

Warsztaty

Materiały do warsztatów znajdują się w podlinkowanym repozytorium

2017/2018

2016/2017

Polecane linki zewnętrzne

Abstrakty prezentacji

Deep Reinforcement Learning for Conversational Agents by Paweł Budzianowski

Bio

Pawel Budzianowski received his B.A. and M.A. degrees from the Faculty of Mathematics and Computer Science at Adam Mickiewicz University in Poznan in 2015. Since then he has been at the University of Cambridge first as an MPhil student reading Machine Learning. Afterwards, he has begun a PhD in the Dialogue Systems Group at the University of Cambridge. His research interests include multi-domain policy management and Bayesian deep learning

Abstract

In spoken dialogue systems, we aim at building automated dialogue agents that can converse with humans. A part of this effort is the policy optimization task, which attempts to find a policy guiding the conversation. This talk will give a gentle introduction to data-driven dialogue systems and address some challenges in training dialogue management module via deep reinforcement learning - namely sample efficiency and stability of learning. We will discuss whether Bayesian approaches can improve stability and efficiency of value-based methods.