A NEW METHOD FOR IMPROVING STUDENT NETWORKS IN COMPUTER VISION

Summary from the TLDR AI newsletter

Deep neural networks have excelled in computer vision, but faster inference times are needed. This paper introduces the Inter-Class Similarity Distillation method and an Adaptive Loss Weighting strategy for better knowledge transfer from a teacher network to a student one.