In computer game-playing, the established method for constructing an evaluation function uses a scalar value computed as a weighted sum of features. This paper advocates the use of partial order evaluation, and describes an ecient new search method called partial order bounding (POB). Previous tree search algorithms using a partial order evaluation have attempted to propagate partially ordered values through the search tree, which leads to many problems in practice, such as the complexity of backing up sets of incomparable evaluations. POB compares partially ordered values only in the leaves of a game tree, and backs up boolean values through the tree. A closely related new algorithm, linear extension partial order bounding (LE-POB), uses a standard scalar alphabeta search with values from a suitably chosen linear extension of the partial order evaluation. As an application, the eectiveness of partial order evaluation is shown in the case of modeling capturing races called semeai in ...

A set of experiments for learning the values of chess pieces is described for the popular chess variants Crazyhouse Chess, Suicide Chess, and Atomic Chess. We follow an established methodology that relies on reinforcement learning from self-games. We attempt to learn piece values and the piecesquare tables for three chess variants. The piece values arrived at, are quite different from those of standard chess, and in several ways surprising, but they generally outperform the values that have been previously used in the literature, and in the implementations of computer players for these games. The results also underline the practical importance of piece-square tables for tactical variants of the game. 1.

Abstract. This paper presents Abalearn, a self-teaching Abalone program capable of automatically reaching an intermediate level of play without needing expert-labeled training examples or deep searches. Our approach is based on a reinforcement learning algorithm that is riskseeking, since defensive players in Abalone tend to never end a game. We extend the risk-sensitive reinforcement learning framework in order to deal with large state spaces and we also propose a set of features that seem relevant for achieving a good level of play. We evaluate our approach using a fixed heuristic opponent as a benchmark, pitting our agents against human players online and comparing samples of our agents at different times of training. 1