{"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/deepsz-a-novel-framework-to-compress-deep","title":"DeepSZ: A Novel Framework to Compress Deep Neural Networks by Using Error-Bounded Lossy Compression","arxiv_id":"1901.09124","date":"2019-01-26","proceeding":null,"authors":["Sian Jin","Sheng Di","Xin Liang","Jiannan Tian","Dingwen Tao","Franck Cappello"],"abstract":"DNNs have been quickly and broadly exploited to improve the data analysis\nquality in many complex science and engineering applications. Today's DNNs are\nbecoming deeper and wider because of increasing demand on the analysis quality\nand more and more complex applications to resolve. The wide and deep DNNs,\nhowever, require large amounts of resources, significantly restricting their\nutilization on resource-constrained systems. Although some network\nsimplification methods have been proposed to address this issue, they suffer\nfrom either low compression ratios or high compression errors, which may\nintroduce a costly retraining process for the target accuracy. In this paper,\nwe propose DeepSZ: an accuracy-loss bounded neural network compression\nframework, which involves four key steps: network pruning, error bound\nassessment, optimization for error bound configuration, and compressed model\ngeneration, featuring a high compression ratio and low encoding time. The\ncontribution is three-fold. (1) We develop an adaptive approach to select the\nfeasible error bounds for each layer. (2) We build a model to estimate the\noverall loss of accuracy based on the accuracy degradation caused by individual\ndecompressed layers. (3) We develop an efficient optimization algorithm to\ndetermine the best-fit configuration of error bounds in order to maximize the\ncompression ratio under the user-set accuracy constraint. Experiments show that\nDeepSZ can compress AlexNet and VGG-16 on the ImageNet by a compression ratio\nof 46X and 116X, respectively, and compress LeNet-300-100 and LeNet-5 on the\nMNIST by a compression ratio of 57X and 56X, respectively, with only up to 0.3%\nloss of accuracy. Compared with other state-of-the-art methods, DeepSZ can\nimprove the compression ratio by up to 1.43X, the DNN encoding performance by\nup to 4.0X (with four Nvidia Tesla V100 GPUs), and the decoding performance by\nup to 6.2X.","url_abs":"http://arxiv.org/abs/1901.09124v2","url_pdf":"http://arxiv.org/pdf/1901.09124v2.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":"deepsz-a-novel-framework-to-compress-deep","repo_url":"https://github.com/szcompressor/DeepSZ","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"caffe2","reach":null}],"tasks":[{"task_slug":"network-pruning","task_name":"Network Pruning"},{"task_slug":"neural-network-compression","task_name":"Neural Network Compression"}],"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":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}