| 摘要 |
Provided is a technology whereby, in relation to machine learning of a parameter used in feature conversion, it is possible to obtain the parameter whereby it is possible to increase task precision, and to make the machine learning more efficient. A feature conversion learning device (1) comprises an approximation unit (5), a loss calculation unit (7), an approximation control unit (6), and a loss control unit (8). The approximation unit (5) takes a feature value that is extracted from a sample pattern and then weighted by a training parameter, assigns that weighted feature value to a variable of a continuous approximation function approximating a step function, and, by doing so, computes an approximated feature value. The loss calculation unit (7) calculates a loss with respect to the task on the basis of the approximated feature value. The approximation control unit (6) controls an approximation precision of the approximation function with respect to the step function such that the approximation function used with the approximation unit (5) approaches the step function according to a decrease in the loss. The loss control unit (8) updates the training parameter such that the loss decreases. |
