RECOGNITION DICTIONARY TRAINING METHOD, SYSTEM, AND PROGRAM

摘要

Disclosed is a recognition dictionary training method, system, and program which allows a computer to function as a recognition dictionary training system, where input vectors other than training data do not cause a decrease in recognition ability. In the recognition dictionary training system, an initial value setting means (11) initializes each parameter against a formula derived from parameter expression in a logistic recurrence framework using multiple heavy-tailed distribution functions. An input means (12) inputs the subset of input vectors for training from the group of training data input vectors. A revised parameter calculation means (13) calculates revised parameter values which decrease the value of the initially established evaluation function. A parameter revision means (14) updates parameters on the basis of said revised values. An end determination means (15) repeats the input means (12) through the parameter revision means (14) until the end is determined to have been reached.