Learning to Play Games in Extensive Form by Valuation

Title: Learning to Play Games in Extensive Form by Valuation

Abstract: This research explores the concept of valuation in games, particularly in extensive form. It proposes a simple learning process where a player's valuation is revised after each play, based on the payoff obtained. The study shows that a player with a winning strategy in a win-lose game can almost surely guarantee a win in a repeated game using this learning process. When multiple payoffs are involved, a more elaborate learning procedure is required. The research suggests that with some perturbations, strategies that are close to subgame perfect equilibrium are played after some time.

Research Question: Can a player improve their performance in games by learning from their own experiences and the outcomes of their moves?

Methodology: The study uses a valuation-based learning process, where a player assigns numeric values to their moves based on their desirability. The player chooses the move with the highest valuation. The valuation is then revised after each play, taking into account the payoff obtained. The research considers a simple revision rule and a more elaborate one for multiple payoffs.

Results: The study shows that a player with a winning strategy in a win-lose game can almost surely guarantee a win in a repeated game using the simple learning process. When multiple payoffs are involved, the more elaborate learning procedure is required to achieve better results. The research also suggests that with some perturbations, strategies that are close to subgame perfect equilibrium are played after some time.

Implications: This research has implications for game theory and artificial intelligence. It suggests that valuation-based learning can be an effective strategy for improving performance in games, particularly in extensive form. It also provides insights into the dynamics of learning in games and the factors that can influence the outcomes.

In conclusion, this research demonstrates that valuation-based learning can be an effective strategy for improving performance in games, particularly in extensive form. It also provides insights into the dynamics of learning in games and the factors that can influence the outcomes.

Link to Article: https://arxiv.org/abs/0201021v1 Authors: arXiv ID: 0201021v1