Reinforcement learning acceleration in multiple goal systems

• Grant number: 07/02279-1
• Support Opportunities: Scholarships in Brazil - Master
• Start date: November 01, 2007
• End date: August 31, 2009
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Carlos Henrique Costa Ribeiro
• Grantee: Helen Cristina de Mattos Senefonte
• Host Institution: Instituto Tecnológico de Aeronáutica (ITA). Ministério da Defesa (Brasil). São José dos Campos , SP, Brazil

Abstract

The goal of this project is to implement and analyze techniques for speeding up reinforcement learning in multiple goal systems. Multiple goal problems can be described in many different ways. The focus here is on cases in which a single agent must learn simultaneously and online many independent sub-tasks resultant from an a priori decomposition of the problem at stake. There is therefore a Distributed Problem Solving instance in that no previous planning occurs, being the agent responsible for the autonomous learning of an action selection process in which a competition between many sub-tasks may happen. The project will involve a formal and empirical analysis based on previous results from the literature, and thereafter we will suggest learning speed up techniques based on heuristics studied and tested in the context of simple goal problems.