The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge


Research Paper #783

Title:Investigating the Behaviour of Q(Lambda)
Authors:Wyatt,J; Hayes,GM; Hallam,JC
Date:Jan 1996
Presented:Presented at the IEE Seminar on Self-Learning Robots
Keywords:
Abstract:$Q(\lambda)$ is an interesting model-free learning algorithm. Unfortunately as originally implemented it is exploration sensitive. We reimplement it using Watkins' rule for correcting eligibility traces, and demonstrate empirically that using these makes the algorithm exploration insensitive once more.
Download:POSTSCRIPT COPY


[Search These Pages] [DAI Home Page] [Comment]