Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386076537> ?p ?o ?g. }
Showing items 1 to 80 of
80
with 100 items per page.
- W4386076537 abstract "Post-training quantization (PTQ) is widely regarded as one of the most efficient compression methods practically, benefitting from its data privacy and low computation costs. We argue that an overlooked problem of oscillation is in the PTQ methods. In this paper, we take the initiative to explore and present a theoretical proof to explain why such a problem is essential in PTQ. And then, we try to solve this problem by introducing a principled and generalized frame-work theoretically. In particular, we first formulate the oscillation in PTQ and prove the problem is caused by the difference in module capacity. To this end, we define the module capacity (ModCap) under data-dependent and data-free scenarios, where the differentials between adjacent modules are used to measure the degree of oscillation. The problem is then solved by selecting top-k differentials, in which the corresponding modules are jointly optimized and quantized. Extensive experiments demonstrate that our method successfully reduces the performance drop and is generalized to different neural networks and PTQ methods. For example, with 2/4 bit ResNet-50 quantization, our method surpasses the previous state-of-the-art method by 1.9%. It becomes more significant on small model quantization, e.g. surpasses BRECQ method by 6.61% on MobileNetV2 × 0.5." @default.
- W4386076537 created "2023-08-23" @default.
- W4386076537 creator A5000389309 @default.
- W4386076537 creator A5002247802 @default.
- W4386076537 creator A5016080094 @default.
- W4386076537 creator A5034191292 @default.
- W4386076537 creator A5035873669 @default.
- W4386076537 creator A5054226277 @default.
- W4386076537 creator A5054612059 @default.
- W4386076537 creator A5070812231 @default.
- W4386076537 creator A5089480544 @default.
- W4386076537 date "2023-06-01" @default.
- W4386076537 modified "2023-09-27" @default.
- W4386076537 title "Solving Oscillation Problem in Post-Training Quantization Through a Theoretical Perspective" @default.
- W4386076537 cites W2065219958 @default.
- W4386076537 cites W2117539524 @default.
- W4386076537 cites W2194775991 @default.
- W4386076537 cites W2618530766 @default.
- W4386076537 cites W2963163009 @default.
- W4386076537 cites W2963521187 @default.
- W4386076537 cites W2964350391 @default.
- W4386076537 cites W2981383995 @default.
- W4386076537 cites W2982479999 @default.
- W4386076537 cites W3004061291 @default.
- W4386076537 cites W3034571205 @default.
- W4386076537 cites W3034940165 @default.
- W4386076537 cites W3035232708 @default.
- W4386076537 cites W3134119092 @default.
- W4386076537 cites W3146523516 @default.
- W4386076537 cites W3176068477 @default.
- W4386076537 cites W3202442802 @default.
- W4386076537 cites W3205652345 @default.
- W4386076537 cites W4226437046 @default.
- W4386076537 cites W4312559913 @default.
- W4386076537 cites W4312690709 @default.
- W4386076537 cites W4312950795 @default.
- W4386076537 cites W4318954855 @default.
- W4386076537 doi "https://doi.org/10.1109/cvpr52729.2023.00768" @default.
- W4386076537 hasPublicationYear "2023" @default.
- W4386076537 type Work @default.
- W4386076537 citedByCount "0" @default.
- W4386076537 crossrefType "proceedings-article" @default.
- W4386076537 hasAuthorship W4386076537A5000389309 @default.
- W4386076537 hasAuthorship W4386076537A5002247802 @default.
- W4386076537 hasAuthorship W4386076537A5016080094 @default.
- W4386076537 hasAuthorship W4386076537A5034191292 @default.
- W4386076537 hasAuthorship W4386076537A5035873669 @default.
- W4386076537 hasAuthorship W4386076537A5054226277 @default.
- W4386076537 hasAuthorship W4386076537A5054612059 @default.
- W4386076537 hasAuthorship W4386076537A5070812231 @default.
- W4386076537 hasAuthorship W4386076537A5089480544 @default.
- W4386076537 hasConcept C11413529 @default.
- W4386076537 hasConcept C28855332 @default.
- W4386076537 hasConcept C33923547 @default.
- W4386076537 hasConcept C41008148 @default.
- W4386076537 hasConcept C80444323 @default.
- W4386076537 hasConceptScore W4386076537C11413529 @default.
- W4386076537 hasConceptScore W4386076537C28855332 @default.
- W4386076537 hasConceptScore W4386076537C33923547 @default.
- W4386076537 hasConceptScore W4386076537C41008148 @default.
- W4386076537 hasConceptScore W4386076537C80444323 @default.
- W4386076537 hasFunder F4320321001 @default.
- W4386076537 hasFunder F4320335777 @default.
- W4386076537 hasFunder F4320336125 @default.
- W4386076537 hasLocation W43860765371 @default.
- W4386076537 hasOpenAccess W4386076537 @default.
- W4386076537 hasPrimaryLocation W43860765371 @default.
- W4386076537 hasRelatedWork W1564189257 @default.
- W4386076537 hasRelatedWork W2043855256 @default.
- W4386076537 hasRelatedWork W2100220075 @default.
- W4386076537 hasRelatedWork W2136607426 @default.
- W4386076537 hasRelatedWork W2166659185 @default.
- W4386076537 hasRelatedWork W2168553558 @default.
- W4386076537 hasRelatedWork W2386767533 @default.
- W4386076537 hasRelatedWork W2541477304 @default.
- W4386076537 hasRelatedWork W3011722122 @default.
- W4386076537 hasRelatedWork W2185799413 @default.
- W4386076537 isParatext "false" @default.
- W4386076537 isRetracted "false" @default.
- W4386076537 workType "article" @default.