EARLY INFERENCE BASED DEEP LEARNING DEVICE AND METHOD THEREOF

摘要

The present invention relates to an early inference based deep learning device capable of determining the effective position and number of additional output layers with low complexity and connecting an output layer only between relatively high accuracy layers by using the same, and a method thereof. The early inference based deep learning device according to the present invention can comprise: a multi-output collection part collecting intermediate output values in all layers of a pre-learned deep learning network; an output layer position inference part clustering the collected intermediate output values for each layer to infer an additional output layer position; and a network learning part that forms a new deep learning network by connecting an output layer to the inferred additional output layer position, and learns the formed new deep learning network.;COPYRIGHT KIPO 2020