{"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/proxquant-quantized-neural-networks-via","title":"ProxQuant: Quantized Neural Networks via Proximal Operators","arxiv_id":"1810.00861","date":"2018-10-01","proceeding":"ICLR 2019 5","authors":["Yu Bai","Yu-Xiang Wang","Edo Liberty"],"abstract":"To make deep neural networks feasible in resource-constrained environments\n(such as mobile devices), it is beneficial to quantize models by using\nlow-precision weights. One common technique for quantizing neural networks is\nthe straight-through gradient method, which enables back-propagation through\nthe quantization mapping. Despite its empirical success, little is understood\nabout why the straight-through gradient method works.\n  Building upon a novel observation that the straight-through gradient method\nis in fact identical to the well-known Nesterov's dual-averaging algorithm on a\nquantization constrained optimization problem, we propose a more principled\nalternative approach, called ProxQuant, that formulates quantized network\ntraining as a regularized learning problem instead and optimizes it via the\nprox-gradient method. ProxQuant does back-propagation on the underlying\nfull-precision vector and applies an efficient prox-operator in between\nstochastic gradient steps to encourage quantizedness. For quantizing ResNets\nand LSTMs, ProxQuant outperforms state-of-the-art results on binary\nquantization and is on par with state-of-the-art on multi-bit quantization. For\nbinary quantization, our analysis shows both theoretically and experimentally\nthat ProxQuant is more stable than the straight-through gradient method (i.e.\nBinaryConnect), challenging the indispensability of the straight-through\ngradient method and providing a powerful alternative.","url_abs":"http://arxiv.org/abs/1810.00861v3","url_pdf":"http://arxiv.org/pdf/1810.00861v3.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":"proxquant-quantized-neural-networks-via","repo_url":"https://github.com/allenbai01/ProxQuant","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"quantization","task_name":"Quantization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1810.00861","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1810.00861"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/allenbai01/ProxQuant","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran_draft_wrong":1,"unverified":15},"by_repo_kind":{"official":{"samples":16,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"fac5364e2f53c6db","entry":"conv3x3","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"models/resnet.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/models/resnet.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"fac5364e2f53c6db"}},{"code_sha256_prefix":"87e04701edce84bc","entry":"Binarize","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"models/binarized_modules.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/models/binarized_modules.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"87e04701edce84bc"}},{"code_sha256_prefix":"11d4ff1157dd777a","entry":"Binaryconv3x3","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"models/resnet_binary.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/models/resnet_binary.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"11d4ff1157dd777a"}},{"code_sha256_prefix":"389e9d00ee17c1e7","entry":"Quantize","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"models/binarized_modules.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/models/binarized_modules.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"389e9d00ee17c1e7"}},{"code_sha256_prefix":"ce88743b4a4532aa","entry":"find_B_numpy","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"bst.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/bst.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ce88743b4a4532aa"}},{"code_sha256_prefix":"ea137766123e8a10","entry":"find_B_torch","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"bst.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/bst.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ea137766123e8a10"}},{"code_sha256_prefix":"172df2baa8f002c3","entry":"forward","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"main_binary_reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/main_binary_reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"172df2baa8f002c3"}},{"code_sha256_prefix":"3b6b395f11bd2a5e","entry":"get_dataset","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"data.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/data.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3b6b395f11bd2a5e"}},{"code_sha256_prefix":"2bdeac768f5f8258","entry":"if_binary","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2bdeac768f5f8258"}},{"code_sha256_prefix":"489f980022491193","entry":"if_binary_lm","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"489f980022491193"}},{"code_sha256_prefix":"5efcd1e9ae48efb9","entry":"if_binary_tern","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5efcd1e9ae48efb9"}},{"code_sha256_prefix":"f407103e88c43c4d","entry":"inception_preproccess","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"preprocess.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/preprocess.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f407103e88c43c4d"}},{"code_sha256_prefix":"65a7404ee8437540","entry":"pad_random_crop","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"preprocess.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/preprocess.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"65a7404ee8437540"}},{"code_sha256_prefix":"7e8728eb31828d77","entry":"scale_crop","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"preprocess.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/preprocess.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"7e8728eb31828d77"}},{"code_sha256_prefix":"a6c968b28755b58c","entry":"train","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"main_binary_reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/main_binary_reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"a6c968b28755b58c"}},{"code_sha256_prefix":"6604ddcbff9362bd","entry":"validate","repo":"allenbai01/ProxQuant","repo_kind":"official","path":"main_binary_reg.py","file_url":"https://github.com/allenbai01/ProxQuant/blob/HEAD/main_binary_reg.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"6604ddcbff9362bd"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}