Global Utilities

Research - Current Postgraduates - Details

Department of Computer Science & Computer Engineering

Darbyshire, Paul
Course: PhD
Research Title/Topic: Using Communication to Achieve Group Goals Beyond Individual Capabilities
Supervisor: Assoc. Prof. Justin Wang
Description:
Reinforcement learning has received a lot of attention by researchers when implementing intelligence within multi-agent systems, due to the nature of paradigm. However, trials have indicated that a team of reinforcement learning agents, when teamed against an identical sized and equipped team of agents utilizing a standard agent control algorithm do not learn fast enough to survive. If the agents communicate during the learning process, the learning rate can be accelerated sufficiently to increase the chance of survival, and triumph over the opposing team. This research looks at the nature and significance of the communication between the learning entities, and the ability to accelerate the learning process through communication. Accelerated learning within such a system could enable the system as a whole to achieve a goal that is beyond the ability of the individual entities.
Content Approved by: Head of School
Page maintained by: Applications Programmer
Last Updated: 14 October, 2009