The problem of induction and machine learning

摘要: Are we justified in inferring a general rule from observations that frequently confirm it? This is the usual statement of problem induction. The present paper argues this question relevant for understanding Machine Learning, but insufficient. Research Learning has prompted another, more fundamental question: number possible rules grows exponentially with size examples, and many them are somehow confirmed by data - how to choose effectively some have good chances being predictive? We analyze if approached standard accounts induction show difficulties present. Finally, suggest Explanation-based approach related methods knowledge intensive could be partial solution these problems, help valid new perspective.