| CPC G06N 3/063 (2013.01) [G06F 18/214 (2023.01); G06F 18/2431 (2023.01); G06F 18/2451 (2023.01); G06F 18/2453 (2023.01); G06N 3/08 (2013.01); G06N 20/10 (2019.01)] | 10 Claims |
|
10. A quantization parameter optimization device that determines a quantization parameter that is a weight parameter in a neural network having been quantized, the quantization parameter optimization device comprising:
a cost function determiner that determines a cost function in which a regularization term is added to an error function, the regularization term being a function of a quantization error that is an error between the weight parameter and the quantization parameter, the error function being a function for determining an error between an output value of the neural network and a corresponding teaching data value;
an updater that updates the quantization parameter by use of the cost function; and
a quantization parameter determiner that determines, as an optimized quantization parameter of a quantization neural network, the quantization parameter with which a function value derived from the cost function satisfies a predetermined condition, the optimized quantization parameter being obtained as a result of the updater repeating updating the quantization parameter, the quantization neural network being the neural network, the weight parameter of which has been quantized,
wherein the function value derived from the regularization term and an inference accuracy of the quantization neural network are negatively correlated.
|