{"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/syq-learning-symmetric-quantization-for","title":"SYQ: Learning Symmetric Quantization For Efficient Deep Neural Networks","arxiv_id":"1807.00301","date":"2018-07-01","proceeding":"CVPR 2018 6","authors":["Julian Faraone","Nicholas Fraser","Michaela Blott","Philip H. W. Leong"],"abstract":"Inference for state-of-the-art deep neural networks is computationally\nexpensive, making them difficult to deploy on constrained hardware\nenvironments. An efficient way to reduce this complexity is to quantize the\nweight parameters and/or activations during training by approximating their\ndistributions with a limited entry codebook. For very low-precisions, such as\nbinary or ternary networks with 1-8-bit activations, the information loss from\nquantization leads to significant accuracy degradation due to large gradient\nmismatches between the forward and backward functions. In this paper, we\nintroduce a quantization method to reduce this loss by learning a symmetric\ncodebook for particular weight subgroups. These subgroups are determined based\non their locality in the weight matrix, such that the hardware simplicity of\nthe low-precision representations is preserved. Empirically, we show that\nsymmetric quantization can substantially improve accuracy for networks with\nextremely low-precision weights and activations. We also demonstrate that this\nrepresentation imposes minimal or no hardware implications to more\ncoarse-grained approaches. Source code is available at\nhttps://www.github.com/julianfaraone/SYQ.","url_abs":"http://arxiv.org/abs/1807.00301v1","url_pdf":"http://arxiv.org/pdf/1807.00301v1.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":"syq-learning-symmetric-quantization-for","repo_url":"https://github.com/julianfaraone/SYQ","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"quantization","task_name":"Quantization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1807.00301","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}