Method and apparatus for on-device continuous learning of neural networks analyzing input data for use in smartphones, drones, ships or military purposes, and method and apparatus for testing neural networks learned thereby

摘要

PROBLEM TO BE SOLVED: To provide a method for performing on-device continuous learning of a neural network for analyzing input data. A learning device (100) samples new data so as to have a preset first volume and uses a previously learned original data generator network to synthesize pre-combined data (first data) corresponding to a k-dimensional random vector. 1 pre-synthesis data corresponds to the previous data used to previously train the original data generator network) and outputs the pre-synthesis data to the second volume, Is used to generate output information corresponding to the batch using a neural network. The method is performed by GAN (general adversarial network), online learning, or the like. [Selection diagram] Figure 2