| 摘要 |
PURPOSE:To decide whether a stochastic rule by which inspection data is generated is the same as a stochastic rule by which training data is generated, with error probability being as small as possible even from a small number of data, when the training data and the inspection data are given. CONSTITUTION:In a step 11, description length to finite pieces of models is calculated with respect to training data, and in a step 12, the model for giving the minimum description length and a real number value parameter which follows it are determined. In a step 13, a value of a discriminant function is calculated by using a difference between logarithmic likelihood of inspection data to the model and the real number value parameter, and logarithmic likelihood of a known stochastic rule. In a step 14, when the discriminant function is positive, it is decided that the data is generated in conformity with inspection data P1, a signal Y for showing it, and when the discriminant function is negative, it is decided that the inspection data is generated in conformity with P2, and a signal N for showing it are outputted as a result of decision, respectively. |
