Matches in SemOpenAlex for { <https://semopenalex.org/work/W2904640605> ?p ?o ?g. }
- W2904640605 abstract "Deep neural networks have achieved state-of-the-art accuracies in a wide range of computer vision, speech recognition, and machine translation tasks. However the limits of memory bandwidth and computational power constrain the range of devices capable of deploying these modern networks. To address this problem, we propose SQuantizer, a new training method that jointly optimizes for both sparse and low-precision neural networks while maintaining high accuracy and providing a high compression rate. This approach brings sparsification and low-bit quantization into a single training pass, employing these techniques in an order demonstrated to be optimal. Our method achieves state-of-the-art accuracies using 4-bit and 2-bit precision for ResNet18, MobileNet-v2 and ResNet50, even with high degree of sparsity. The compression rates of 18x for ResNet18 and 17x for ResNet50, and 9x for MobileNet-v2 are obtained when SQuantizing both weights and activations within 1% and 2% loss in accuracy for ResNets and MobileNet-v2 respectively. An extension of these techniques to object detection also demonstrates high accuracy on YOLO-v3. Additionally, our method allows for fast single pass training, which is important for rapid prototyping and neural architecture search techniques. Finally extensive results from this simultaneous training approach allows us to draw some useful insights into the relative merits of sparsity and quantization." @default.
- W2904640605 created "2018-12-22" @default.
- W2904640605 creator A5028464094 @default.
- W2904640605 creator A5030884761 @default.
- W2904640605 creator A5076399105 @default.
- W2904640605 date "2018-12-20" @default.
- W2904640605 modified "2023-09-24" @default.
- W2904640605 title "SQuantizer: Simultaneous Learning for Both Sparse and Low-precision Neural Networks" @default.
- W2904640605 cites W1861492603 @default.
- W2904640605 cites W2117539524 @default.
- W2904640605 cites W2181101938 @default.
- W2904640605 cites W2194775991 @default.
- W2904640605 cites W2242818861 @default.
- W2904640605 cites W2285660444 @default.
- W2904640605 cites W2469490737 @default.
- W2904640605 cites W2515385951 @default.
- W2904640605 cites W2553303224 @default.
- W2904640605 cites W2554302513 @default.
- W2904640605 cites W2560017826 @default.
- W2904640605 cites W2593245696 @default.
- W2904640605 cites W2625457103 @default.
- W2904640605 cites W2707890836 @default.
- W2904640605 cites W2737244778 @default.
- W2904640605 cites W2754084392 @default.
- W2904640605 cites W2769900398 @default.
- W2904640605 cites W2786054724 @default.
- W2904640605 cites W2786771851 @default.
- W2904640605 cites W2796347433 @default.
- W2904640605 cites W2799238091 @default.
- W2904640605 cites W2809624076 @default.
- W2904640605 cites W2886847195 @default.
- W2904640605 cites W2886851211 @default.
- W2904640605 cites W2899067310 @default.
- W2904640605 cites W2899771611 @default.
- W2904640605 cites W2915589364 @default.
- W2904640605 cites W2921967969 @default.
- W2904640605 cites W2951433694 @default.
- W2904640605 cites W2963122961 @default.
- W2904640605 cites W2963125010 @default.
- W2904640605 cites W2963163009 @default.
- W2904640605 cites W2963363373 @default.
- W2904640605 cites W2963446712 @default.
- W2904640605 cites W2963521187 @default.
- W2904640605 cites W2963674932 @default.
- W2904640605 cites W2963723401 @default.
- W2904640605 cites W2963769126 @default.
- W2904640605 cites W2963981420 @default.
- W2904640605 cites W2964233199 @default.
- W2904640605 cites W2964299589 @default.
- W2904640605 cites W992687842 @default.
- W2904640605 hasPublicationYear "2018" @default.
- W2904640605 type Work @default.
- W2904640605 sameAs 2904640605 @default.
- W2904640605 citedByCount "4" @default.
- W2904640605 countsByYear W29046406052019 @default.
- W2904640605 countsByYear W29046406052021 @default.
- W2904640605 crossrefType "posted-content" @default.
- W2904640605 hasAuthorship W2904640605A5028464094 @default.
- W2904640605 hasAuthorship W2904640605A5030884761 @default.
- W2904640605 hasAuthorship W2904640605A5076399105 @default.
- W2904640605 hasConcept C108583219 @default.
- W2904640605 hasConcept C113775141 @default.
- W2904640605 hasConcept C11413529 @default.
- W2904640605 hasConcept C154945302 @default.
- W2904640605 hasConcept C159985019 @default.
- W2904640605 hasConcept C192562407 @default.
- W2904640605 hasConcept C204323151 @default.
- W2904640605 hasConcept C2776257435 @default.
- W2904640605 hasConcept C28855332 @default.
- W2904640605 hasConcept C2984842247 @default.
- W2904640605 hasConcept C31258907 @default.
- W2904640605 hasConcept C41008148 @default.
- W2904640605 hasConcept C50644808 @default.
- W2904640605 hasConceptScore W2904640605C108583219 @default.
- W2904640605 hasConceptScore W2904640605C113775141 @default.
- W2904640605 hasConceptScore W2904640605C11413529 @default.
- W2904640605 hasConceptScore W2904640605C154945302 @default.
- W2904640605 hasConceptScore W2904640605C159985019 @default.
- W2904640605 hasConceptScore W2904640605C192562407 @default.
- W2904640605 hasConceptScore W2904640605C204323151 @default.
- W2904640605 hasConceptScore W2904640605C2776257435 @default.
- W2904640605 hasConceptScore W2904640605C28855332 @default.
- W2904640605 hasConceptScore W2904640605C2984842247 @default.
- W2904640605 hasConceptScore W2904640605C31258907 @default.
- W2904640605 hasConceptScore W2904640605C41008148 @default.
- W2904640605 hasConceptScore W2904640605C50644808 @default.
- W2904640605 hasLocation W29046406051 @default.
- W2904640605 hasOpenAccess W2904640605 @default.
- W2904640605 hasPrimaryLocation W29046406051 @default.
- W2904640605 hasRelatedWork W2580414508 @default.
- W2904640605 hasRelatedWork W2594107175 @default.
- W2904640605 hasRelatedWork W2786771851 @default.
- W2904640605 hasRelatedWork W2896988829 @default.
- W2904640605 hasRelatedWork W2909712920 @default.
- W2904640605 hasRelatedWork W2931497102 @default.
- W2904640605 hasRelatedWork W2982041622 @default.
- W2904640605 hasRelatedWork W2991002474 @default.
- W2904640605 hasRelatedWork W3007739908 @default.
- W2904640605 hasRelatedWork W3023803297 @default.
- W2904640605 hasRelatedWork W3027574712 @default.