Hypothesis Filtering: A Practical Approach to Reliable Learning

摘要: Abstract This paper shows how to take virtually any learning algorithm and turn it into one which is accurate reliable an arbitrary degree. The transformation accomplished by appending a filter that performs statistical test the algorithm. only outputs hypotheses pass test. takes time polynomial in desired accuracy reliability levels independent of complexity its domain. Distribution-free theory used prove works. We describe application concept learning, SE, system learns control knowledge clustering data. significance hypothesis filtering two fold. First, filters may be evaluate compare performance wide range algorithms. Second, applying inductive algorithms demonstrates can made arbitrarily reliable, provably so.