| 摘要 |
<p>A learning method for adjusting control parameters THETA of a learning apparatus in which input data is converted into output data by utilizing the control parameters THETA and teaching signal data consists of steps of defining undesirable output data as a type of the teaching signal data, defining a loss function r = exp(-x<2>) of which a value decreases along with the increase of a difference x between the undesirable output data and the output data, providing both the input data and the undesirable output data to the learning apparatus, calculating a value of the output data by converting a value of the input data by utilizing values of the control parameters THETA , calculating a value of the loss function r by utilizing both the value of the output data and a value of the undesirable output data, iteratively calculating the value of the loss function r by renewing the values of the control parameters THETA to decrease the value of the loss function r, and adjusting the control parameters THETA of the learning apparatus to newest values of the control parameters THETA when the value of the loss function is less than the prescribed value. <IMAGE></p> |
