Method, Device and System for Running Deep-Learning Model Separately

摘要

The method for performing separation of a deep learning model according to an embodiment of the present invention includes the steps of: a client setting original data as an input value and applying a first function to output intermediate data, wherein the client transmits the intermediate data to a server. Step, the server receiving the intermediate data, and the server setting the intermediate data as an input value and applying a second function to output result data, the first function and the second function A parent function trained with a deep learning model is constructed, and the first function and the second function are partial functions obtained by dividing the parent function by two based on a predetermined criterion. According to an embodiment of the present invention, it is possible to prevent the third party from obtaining important information such as personal information even if a third party intrudes into the network and steals data while providing an online service using a deep learning model. .