2017/07/01 by Yoonho Boo, Boo, Yoonho, Wonyong Sung +1
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Sparse and Compressive Sensing Techniques #cs.CV
paper · pdf · doi:10.48550/arxiv.1707.03684
This paper is accepted in SIPS 2017
arxiv created 2017/07/01 · openalex publication_date 2017/07/01 · arxiv updated 2017/07/13 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Deep neural networks (DNNs) usually demand a large amount of operations for real-time inference. Especially, fully-connected layers contain a large number of weights, thus they usually need many off-chip memory accesses for inference. We propose a weight compression method for deep neural networks, which allows values of +1 or -1 only at predetermined positions of the weights so that decoding using a table can be conducted easily. For example, the structured sparse (8,2) coding allows at most two non-zero values among eight weights. This method not only enables multiplication-free DNN implementations but also compresses the weight storage by up to x32 compared to floating-point networks. Weight distribution normalization and gradual pruning techniques are applied to mitigate the performance degradation. The experiments are conducted with fully-connected deep neural networks and convolutional neural networks.