PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Challenges and Opportunities for Reinforcement Learning in Human Computer Interaction Systems
Satinder Baveja
In: Machine Learning Meets the User Interface, 12 December 2003, Whistler, Canada.


Many human-computer interaction (HCI) systems are sequential interaction systems in which the designer has incomplete and uncertain knowledge about the system's environment and in which user-feedback is impoverished, noisy, and delayed in time. These are precisely the sort of problems reinforcement learning (RL) methods are good at solving. In this talk, I will discuss the opportunities and challenges facing the use of RL as a rigorous design principle for HCI, and illustrate my arguments using examples from 3 simple RL-based HCI systems that I have helped build: an adaptive spoken-dialogue system, an interactive software agent in an online community, and most recently an adaptive reminder system in a cognitive orthotic domain.

EPrint Type:Conference or Workshop Item (Talk)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:User Modelling for Computer Human Interaction
ID Code:70
Deposited By:Steve Gunn
Deposited On:12 May 2004