Cost-Strategy Recognition method for behavior learning: Two players learning simultaneously
Conference Paper
Overview
Identity
View All
Overview
abstract
The Cost-Strategy Recognition method of behavior learning is introduced. A player formulates a behavior model for its opponent by identifying a model structure and then performing parameter identification for that model using the observed behavior of the opponent. If the opponent is not learning and is just playing a fixed strategy, the behavior model of the opponent is also fixed and has to be determined once during the game. However, when the opponent is learning during the game, its behavior changes with time. In such a situation, the player has to perform parameter identification continuously during the game. A learning algorithm based on the Cost-Strategy Recognition (CSR) method is developed that allows the player to continuously update the behavior model as new measurements are made available. Using the developed algorithm, games are simulated in which both players are simultaneously learning each other's behavior.
