Learning to Act: Qualitative Learning of Deterministic Action Models

Thomas Bolander
Technical University of Denmark (DTU), Copenhagen

3:15pm-4:15pm, 1 November 2016
EM G.44

Abstract

In this talk we address the problem of learnability of action models in the context of dynamic epistemic logic. Dynamic epistemic logic is a very expressive formalism for reasoning about (higher-order) knowledge of agents, and for reasoning about the dynamics of such knowledge under the execution of actions. Dynamic epistemic logic provides a very expressive formalism for epistemic planning: planning in which the agents are enriched with the ability to do (higher-order) reasoning about their own knowledge and ignorance, and the knowledge and ignorance of other agents. The ultimate goal of the current work is to integrate learning of actions via observations into epistemic planning. We consider two basic learnability criteria in our setting: finite identifiability (conclusively inferring the appropriate action model in finite time) and identifiability in the limit (inconclusive convergence to the right action model). We show that deterministic actions are finitely identifiable, while arbitrary (non-deterministic) actions require more learning power---they are identifiable in the limit. We then move on to a particular learning method, i.e., learning via update, which proceeds via restriction of a space of events within a learning-specific action model. We show how this method can be adapted to learn conditional and non-conditional action models.

This is joint work with Nina Gierasimczuk.

Bio

Thomas Bolander is an associate professor at the Technical University of Denmark (DTU) in Copenhagen. His research interests are include logic, artificial intelligence, social intelligence, multi-agent systems and automated planning. Of special interest is the modelling of social phenomena and social intelligence with the aim of creating computer systems that can interact intelligently with humans and other computer systems. His recent research focus has been on epistemic planning: enriching the theories of automated planning with the powerful and expressive concepts and structures from dynamic epistemic logic.

Website: http://www.imm.dtu.dk/~tobo/

Host: Ron Petrick