Binary neural networks: A survey

• Published in: Pattern recognition Vol. 105; p. 107281
• Main Authors: Qin, Haotong, Gong, Ruihao, Liu, Xianglong, Bai, Xiao, Song, Jingkuan, Sebe, Nicu
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2020
• Subjects: Binary neural network; Deep learning; Model acceleration; Model compression; Network quantization; Deep learning; Model acceleration; Binary neural network; Network quantization; Model compression
• DOI: 10.1016/j.patcog.2020.107281
• ISSN: 0031-3203

•We summarize the binary neural network methods and categorize them into the naive binarization and the optimized binarization.•The binary neural networks are mainly optimized using techniques including minimizing quantization error, improving the loss function, and reducing the gradient error.•We also discuss the hardware-friendly methods and the useful tricks of training binary neural networks.•We present the common datasets and network structures of evaluation, and compare the performance on different tasks.•We conclude and point out the future research trends. The binary neural network, largely saving the storage and computation, serves as a promising technique for deploying deep models on resource-limited devices. However, the binarization inevitably causes severe information loss, and even worse, its discontinuity brings difficulty to the optimization of the deep network. To address these issues, a variety of algorithms have been proposed, and achieved satisfying progress in recent years. In this paper, we present a comprehensive survey of these algorithms, mainly categorized into the native solutions directly conducting binarization, and the optimized ones using techniques like minimizing the quantization error, improving the network loss function, and reducing the gradient error. We also investigate other practical aspects of binary neural networks such as the hardware-friendly design and the training tricks. Then, we give the evaluation and discussions on different tasks, including image classification, object detection and semantic segmentation. Finally, the challenges that may be faced in future research are prospected.