Computer Science

News and events

25 April 2012

Three approaches to the evolution of intelligent agents

Presented By Dr Alastair Channon, Lecturer, School of Computing and Mathematics, Keele University
  • 3.00pm
  • N112, Haslegrave Building

About this event

Abstract: The conventional approach to evolving intellgent agents is to start with a random population and then, in each generation, select and breed those agents that perform well on a specific task. Success in this appraoch is limited to those tasks for which an evolutionary path can be found from a random or primordial soup to a sufficient behaviour.

This presentation will give an overview of the three leading approaches being taken to overcome this limitation: first scaffolding selection with increasingly complex tasks, second searching for novel behaviours, and third seeding evolution by artificial selection with the results of (long-term or open-ended) evolution by natural selection.

Examples of evolved agents will be shown in simple two-and three-dimensional environments.