Neural network on-device continuous learning method and apparatus for analyzing input data by optimized sampling of training image for smartphone, drone, ship or military purpose, and test method and apparatus using the same

摘要

PROBLEM TO BE SOLVED: To prevent invasion of privacy, optimize resources such as storage, and optimize a training image sampling process in on-device continuous learning of a neural network that analyzes input data. SOLUTION: In the learning device 100, new data is uniform-sampled so as to have a first volume, and with a boosting network, the k-dimensional random vector corresponds to the k-dimensional modified vector and the previous data used for learning. The process of outputting the pre-combination data of the second volume corresponding to the above is repeated, and a step 130 of generating a batch currently used for learning and a neural network to generate output information corresponding to the first batch. Step 125 is included. [Selection diagram] Figure 2