{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/towards-effective-low-bitwidth-convolutional","title":"Towards Effective Low-bitwidth Convolutional Neural Networks","arxiv_id":"1711.00205","date":"2017-11-01","proceeding":"CVPR 2018 6","authors":["Bohan Zhuang","Chunhua Shen","Mingkui Tan","Lingqiao Liu","Ian Reid"],"abstract":"This paper tackles the problem of training a deep convolutional neural\nnetwork with both low-precision weights and low-bitwidth activations.\nOptimizing a low-precision network is very challenging since the training\nprocess can easily get trapped in a poor local minima, which results in\nsubstantial accuracy loss. To mitigate this problem, we propose three\nsimple-yet-effective approaches to improve the network training. First, we\npropose to use a two-stage optimization strategy to progressively find good\nlocal minima. Specifically, we propose to first optimize a net with quantized\nweights and then quantized activations. This is in contrast to the traditional\nmethods which optimize them simultaneously. Second, following a similar spirit\nof the first method, we propose another progressive optimization approach which\nprogressively decreases the bit-width from high-precision to low-precision\nduring the course of training. Third, we adopt a novel learning scheme to\njointly train a full-precision model alongside the low-precision one. By doing\nso, the full-precision model provides hints to guide the low-precision model\ntraining. Extensive experiments on various datasets ( i.e., CIFAR-100 and\nImageNet) show the effectiveness of the proposed methods. To highlight, using\nour methods to train a 4-bit precision network leads to no performance decrease\nin comparison with its full-precision counterpart with standard network\narchitectures ( i.e., AlexNet and ResNet-50).","url_abs":"http://arxiv.org/abs/1711.00205v2","url_pdf":"http://arxiv.org/pdf/1711.00205v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"towards-effective-low-bitwidth-convolutional","repo_url":"https://github.com/nowgood/QuantizeCNNModel","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"towards-effective-low-bitwidth-convolutional","repo_url":"https://github.com/bohanzhuang/Towards-Effective-Low-bitwidth-Convolutional-Neural-Networks","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"quantization","task_name":"Quantization"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"local-response-normalization","method_name":"Local Response Normalization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.00205","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}