Based on the Collective Learning
Theory [Bock 1993]
Collective Learning Systems [Fall 2004] -
Professor Mark Happel
Collective Learning Automata usually use a Static/Uniform
compensation policy to update their STM. Therefore the amount of knowledge the
CLA has gathered in the STM is generally ignored. Additionally, even if the
responses chosen by a CLA are not equally responsible for the final outcome, a
Uniform update policy equally rewards/penalizes all of the weights.
In this paper the performance of a CLA with a Progressive/Discriminant
compensation policy has been compared to a Static/Uniform policy.
A CLA was implemented to learn how to play a game called “Connecting-the-Dots”.
The performance of the CLA with each form of compensation/update policy was
calculated and analyzed.
The conclusions show that a Progressive compensation policy or a Discriminant
update policy can dramatically improve the learning rate of a collective