{"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/same-same-but-different-recovering-neural","title":"Same, Same But Different - Recovering Neural Network Quantization Error Through Weight Factorization","arxiv_id":"1902.01917","date":"2019-02-05","proceeding":null,"authors":["Eldad Meller","Alexander Finkelstein","Uri Almog","Mark Grobman"],"abstract":"Quantization of neural networks has become common practice, driven by the\nneed for efficient implementations of deep neural networks on embedded devices.\nIn this paper, we exploit an oft-overlooked degree of freedom in most networks\n- for a given layer, individual output channels can be scaled by any factor\nprovided that the corresponding weights of the next layer are inversely scaled.\nTherefore, a given network has many factorizations which change the weights of\nthe network without changing its function. We present a conceptually simple and\neasy to implement method that uses this property and show that proper\nfactorizations significantly decrease the degradation caused by quantization.\nWe show improvement on a wide variety of networks and achieve state-of-the-art\ndegradation results for MobileNets. While our focus is on quantization, this\ntype of factorization is applicable to other domains such as network-pruning,\nneural nets regularization and network interpretability.","url_abs":"http://arxiv.org/abs/1902.01917v1","url_pdf":"http://arxiv.org/pdf/1902.01917v1.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":"same-same-but-different-recovering-neural","repo_url":"https://github.com/Adamdad/Samesame","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"network-pruning","task_name":"Network Pruning"},{"task_slug":"quantization","task_name":"Quantization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1902.01917","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}