Learning Game Evaluation Functions with a Compound Linear Machine.
Abstract
This paper proposes a structure for a compound linear machine as a solution to the problem of learning in machine game playing. A possible algorithm for training the two machines is involved. An attempt to use a compound machine for choosing chess moves is reported on. Chapter II briefly presents the background concepts in pattern recognition and machine game playing that underlie the work done. Chapter III presents a proposed structure for a compound linear machine that should be capable of learning in game playing. The general rationale for the proposal is presented also. Chapter IV discusses a possible algorithm for training the compound machine proposed. The rationale for each step of the algorithm is discussed. The game of tic-tac-toe is used as an example in explaining each step. Chapter V compares and contrasts the linear machine approach with other approaches to game playing. Chapter VI presents an attempt to apply the proposal and associated training algorithm to the game of chess. Conclusions and recommendations are given in Chapter VII. Appendices contain supportive material and data on work performed. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1980
- Accession Number
- ADA085710
Entities
People
- William Peter Nelson
Organizations
- Air Force Institute of Technology