On combining decisions from multiple expert imitators for performance

摘要: One approach for artificially intelligent agents wishing to maximise some performance metric in a given domain is learn from collection of training data that consists actions or decisions made by expert, an attempt imitate expert's style. We refer this type agent as expert imitator. In paper we investigate whether can be improved combining multiple imitators. particular, two existing approaches decisions. The first combines employing ensemble voting between second dynamically selects the best imitator use at runtime imitators current environment. these computer poker. create limit and no Texas Hold'em determine their using listed above.