Confidence Learning Machine Based on Algorithmic Theory of Randomness and Dissimilarity Description

A confidence learning machine is put forward based on Kolmogorov’s algorithmic theory of randomness and its algorithm is designed It is proved that the computable randomness test function, which is defined through the sample dissimilarity description function, has the same meaning of the non computable randomness test function defined in Kolmogorov’s theory Three types of dissimilarity description functions are designed through describing the dissimilarity of the sample distance, the support degree of the sample on the boundary, and the response of the sample They are used to realize the confidence learning machine algorithm Good results are obtained for the confidence learning machines in the experiment of the recognition of Cleveland heart data and the verification of handwritten signatures