Machine Learning and the foundations of inductive inference

摘要: The problem of valid induction could be stated as follows: are we justified in accepting a given hypothesis on the basis observations that frequently confirm it? present paper argues this question is relevant for understanding Machine Learning, but insufficient. Recent research inductive reasoning has prompted another, more fundamental question: there not just one rule to tested, large number possible rules, and many these somehow confirmed by data — how restrict space hypotheses choose effectively some rules will probably perform well future examples? We analyze if approached standard accounts show difficulties present. Finally, suggest explanation-based learning approach related methods knowledge intensive be, solution, at least tool solving problems.