SYSTEM FOR MEASURING A CONFIDENCE OF AN INFERENCE RESULT FOR AN EARLY INFERENCE OF A STAGED DEEP LEARNING

摘要

A reliability measurement system for early inference in deep learning is disclosed. Reliability measurement system for early inference of deep learning according to an embodiment of the present disclosure includes a plurality of hidden layers that perform inference into at least one classification on elements to be classified, and input the elements into the hidden layer. A learning unit including an input layer and an output layer for outputting final inferred data related to the classification performed on the elements, and between the plurality of hidden layers, provisional classification to the elements and classification probabilities for the provisional classification are an intermediate output layer for outputting the derived hypothetical data, and the highest classification probability among the classification probabilities of the elements included in the inferential data, and the classification probability of other elements except for the element having the highest classification probability and the highest classification and a reliability measuring unit for measuring the reliability of the classification based on the classification probability pattern between the probabilities.