Gerald J. Tesauro
Gerald J. Tesauro’s TD-Gammon work used backgammon as a setting for learning from experience. The program trained a neural network by playing games against itself. TD-Gammon debuted publicly at the 1992 World Cup of Backgammon. It played exhibition games against leading human players. His March 1995 Communications of the ACM article explains TD-Gammon as a neural network that learns a game-position evaluation function through self-play and outcomes.
Timeline
-
1992
Public games against human players
TD-Gammon debuted publicly at the 1992 World Cup of Backgammon. It played exhibition games against leading human players.
-
1995
A published learning account
His March 1995 Communications of the ACM article explains TD-Gammon as a neural network that learns a game-position evaluation function through self-play and outcomes.
-
1995
Studying delayed feedback
The article uses the program to examine reinforcement learning and temporal credit assignment: how earlier decisions can receive learning signals from later outcomes.
-
2013
Research recognition
AAAI elected Gerald J. Tesauro as a Fellow in 2013.
Sources and credits
Photo: ibm.com; The games that helped AI evolve | IBM; creator not stated. License: Not stated by source. Source: ibm.com. Cropped to a square and resized. Ask us to remove or replace this photo.