GENERALIZATION ABILITY IMPROVEMENT METHOD AND DEVICE FOR RANDOM WEIGHT NETWORK

摘要

Disclosed are a generalization ability improvement method and device for a random weight network. According to the method and device, under the premise that a frame structure of the random weight network is not changed, a target sample maximum in uncertainty value is mined from all training samples; a simulation sample approximately identically distributed with the target sample maximum in uncertainty value is generated; and on the basis of the simulation sample, the weight of an output layer of the random weight network is updated iteratively, so that the underlying information of training samples can be actively mined, thereby improving the generalization ability of the random weight network.