Matches in SemOpenAlex for { <https://semopenalex.org/work/W2997609712> ?p ?o ?g. }
- W2997609712 abstract "Quantization is a promising approach for reducing the inference time and memory footprint of neural networks. However, most existing quantization methods require access to the original training dataset for retraining during quantization. This is often not possible for applications with sensitive or proprietary data, e.g., due to privacy and security concerns. Existing zero-shot quantization methods use different heuristics to address this, but they result in poor performance, especially when quantizing to ultra-low precision. Here, we propose ZeroQ , a novel zero-shot quantization framework to address this. ZeroQ enables mixed-precision quantization without any access to the training or validation data. This is achieved by optimizing for a Distilled Dataset, which is engineered to match the statistics of batch normalization across different layers of the network. ZeroQ supports both uniform and mixed-precision quantization. For the latter, we introduce a novel Pareto frontier based method to automatically determine the mixed-precision bit setting for all layers, with no manual search involved. We extensively test our proposed method on a diverse set of models, including ResNet18/50/152, MobileNetV2, ShuffleNet, SqueezeNext, and InceptionV3 on ImageNet, as well as RetinaNet-ResNet50 on the Microsoft COCO dataset. In particular, we show that ZeroQ can achieve 1.71% higher accuracy on MobileNetV2, as compared to the recently proposed DFQ method. Importantly, ZeroQ has a very low computational overhead, and it can finish the entire quantization process in less than 30s (0.5% of one epoch training time of ResNet50 on ImageNet). We have open-sourced the ZeroQ frameworkfootnote{https://github.com/amirgholami/ZeroQ}." @default.
- W2997609712 created "2020-01-10" @default.
- W2997609712 creator A5015689119 @default.
- W2997609712 creator A5033006662 @default.
- W2997609712 creator A5033384149 @default.
- W2997609712 creator A5047285420 @default.
- W2997609712 creator A5050643298 @default.
- W2997609712 creator A5070987734 @default.
- W2997609712 date "2020-01-01" @default.
- W2997609712 modified "2023-10-16" @default.
- W2997609712 title "ZeroQ: A Novel Zero Shot Quantization Framework" @default.
- W2997609712 cites W1690739335 @default.
- W2997609712 cites W1821462560 @default.
- W2997609712 cites W1861492603 @default.
- W2997609712 cites W1999085092 @default.
- W2997609712 cites W2183341477 @default.
- W2997609712 cites W2194775991 @default.
- W2997609712 cites W2233116163 @default.
- W2997609712 cites W2279098554 @default.
- W2997609712 cites W2300242332 @default.
- W2997609712 cites W2405920868 @default.
- W2997609712 cites W2469490737 @default.
- W2997609712 cites W2515385951 @default.
- W2997609712 cites W2560017826 @default.
- W2997609712 cites W2612445135 @default.
- W2997609712 cites W2619096655 @default.
- W2997609712 cites W2786771851 @default.
- W2997609712 cites W2798332427 @default.
- W2997609712 cites W2809624076 @default.
- W2997609712 cites W2884150179 @default.
- W2997609712 cites W2897049076 @default.
- W2997609712 cites W2898422183 @default.
- W2997609712 cites W2900694899 @default.
- W2997609712 cites W2903260438 @default.
- W2997609712 cites W2912168260 @default.
- W2997609712 cites W2949533892 @default.
- W2997609712 cites W2950726407 @default.
- W2997609712 cites W2962298324 @default.
- W2997609712 cites W2962697884 @default.
- W2997609712 cites W2962761403 @default.
- W2997609712 cites W2963114950 @default.
- W2997609712 cites W2963125010 @default.
- W2997609712 cites W2963163009 @default.
- W2997609712 cites W2963273111 @default.
- W2997609712 cites W2963351448 @default.
- W2997609712 cites W2963674932 @default.
- W2997609712 cites W2964228333 @default.
- W2997609712 cites W2964264300 @default.
- W2997609712 cites W2964299589 @default.
- W2997609712 cites W2973061659 @default.
- W2997609712 cites W2976783886 @default.
- W2997609712 cites W2981751377 @default.
- W2997609712 cites W2982041622 @default.
- W2997609712 cites W2989530497 @default.
- W2997609712 cites W2998218113 @default.
- W2997609712 cites W3035460915 @default.
- W2997609712 doi "https://doi.org/10.48550/arxiv.2001.00281" @default.
- W2997609712 hasPublicationYear "2020" @default.
- W2997609712 type Work @default.
- W2997609712 sameAs 2997609712 @default.
- W2997609712 citedByCount "11" @default.
- W2997609712 countsByYear W29976097122019 @default.
- W2997609712 countsByYear W29976097122020 @default.
- W2997609712 countsByYear W29976097122021 @default.
- W2997609712 crossrefType "posted-content" @default.
- W2997609712 hasAuthorship W2997609712A5015689119 @default.
- W2997609712 hasAuthorship W2997609712A5033006662 @default.
- W2997609712 hasAuthorship W2997609712A5033384149 @default.
- W2997609712 hasAuthorship W2997609712A5047285420 @default.
- W2997609712 hasAuthorship W2997609712A5050643298 @default.
- W2997609712 hasAuthorship W2997609712A5070987734 @default.
- W2997609712 hasBestOaLocation W29976097121 @default.
- W2997609712 hasConcept C111919701 @default.
- W2997609712 hasConcept C113775141 @default.
- W2997609712 hasConcept C11413529 @default.
- W2997609712 hasConcept C119857082 @default.
- W2997609712 hasConcept C124101348 @default.
- W2997609712 hasConcept C127705205 @default.
- W2997609712 hasConcept C154945302 @default.
- W2997609712 hasConcept C165696696 @default.
- W2997609712 hasConcept C2776214188 @default.
- W2997609712 hasConcept C28855332 @default.
- W2997609712 hasConcept C38652104 @default.
- W2997609712 hasConcept C41008148 @default.
- W2997609712 hasConcept C74912251 @default.
- W2997609712 hasConceptScore W2997609712C111919701 @default.
- W2997609712 hasConceptScore W2997609712C113775141 @default.
- W2997609712 hasConceptScore W2997609712C11413529 @default.
- W2997609712 hasConceptScore W2997609712C119857082 @default.
- W2997609712 hasConceptScore W2997609712C124101348 @default.
- W2997609712 hasConceptScore W2997609712C127705205 @default.
- W2997609712 hasConceptScore W2997609712C154945302 @default.
- W2997609712 hasConceptScore W2997609712C165696696 @default.
- W2997609712 hasConceptScore W2997609712C2776214188 @default.
- W2997609712 hasConceptScore W2997609712C28855332 @default.
- W2997609712 hasConceptScore W2997609712C38652104 @default.
- W2997609712 hasConceptScore W2997609712C41008148 @default.
- W2997609712 hasConceptScore W2997609712C74912251 @default.
- W2997609712 hasLocation W29976097121 @default.
- W2997609712 hasOpenAccess W2997609712 @default.