FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3000207760> ?p ?o ?g. }

• W3000207760 abstract "Large deep neural network (DNN) models pose the key challenge to energy efficiency due to the significantly higher energy consumption of off-chip DRAM accesses than arithmetic or SRAM operations. It motivates the intensive research on model compression with two main approaches. Weight pruning leverages the redundancy in the number of weights and can be performed in a non-structured, which has higher flexibility and pruning rate but incurs index accesses due to irregular weights, or structured manner, which preserves the full matrix structure with lower pruning rate. Weight quantization leverages the redundancy in the number of bits in weights. Compared to pruning, quantization is much more hardware-friendly, and has become a must-do step for FPGA and ASIC implementations. This paper provides a definitive answer to the question for the first time. First, we build ADMM-NN-S by extending and enhancing ADMM-NN, a recently proposed joint weight pruning and quantization framework. Second, we develop a methodology for fair and fundamental comparison of non-structured and structured pruning in terms of both storage and computation efficiency. Our results show that ADMM-NN-S consistently outperforms the prior art: (i) it achieves 348x, 36x, and 8x overall weight pruning on LeNet-5, AlexNet, and ResNet-50, respectively, with (almost) zero accuracy loss; (ii) we demonstrate the first fully binarized (for all layers) DNNs can be lossless in accuracy in many cases. These results provide a strong baseline and credibility of our study. Based on the proposed comparison framework, with the same accuracy and quantization, the results show that non-structrued pruning is not competitive in terms of both storage and computation efficiency. Thus, we conclude that non-structured pruning is considered harmful. We urge the community not to continue the DNN inference acceleration for non-structured sparsity." @default.
• W3000207760 created "2020-01-23" @default.
• W3000207760 creator A5006570986 @default.
• W3000207760 creator A5007300551 @default.
• W3000207760 creator A5010753766 @default.
• W3000207760 creator A5013938855 @default.
• W3000207760 creator A5016070401 @default.
• W3000207760 creator A5025596795 @default.
• W3000207760 creator A5034466262 @default.
• W3000207760 creator A5036755436 @default.
• W3000207760 creator A5043582832 @default.
• W3000207760 creator A5043676563 @default.
• W3000207760 creator A5047215143 @default.
• W3000207760 creator A5047916979 @default.
• W3000207760 creator A5080407824 @default.
• W3000207760 date "2019-07-03" @default.
• W3000207760 modified "2023-10-03" @default.
• W3000207760 title "Non-Structured DNN Weight Pruning -- Is It Beneficial in Any Platform?" @default.
• W3000207760 cites W1667652561 @default.
• W3000207760 cites W1686810756 @default.
• W3000207760 cites W2000967104 @default.
• W3000207760 cites W2048266589 @default.
• W3000207760 cites W2067523571 @default.
• W3000207760 cites W2094756095 @default.
• W3000207760 cites W2149933564 @default.
• W3000207760 cites W2152839228 @default.
• W3000207760 cites W2163605009 @default.
• W3000207760 cites W2164278908 @default.
• W3000207760 cites W2165698076 @default.
• W3000207760 cites W2194775991 @default.
• W3000207760 cites W2233116163 @default.
• W3000207760 cites W2261808795 @default.
• W3000207760 cites W2267635276 @default.
• W3000207760 cites W2276486856 @default.
• W3000207760 cites W2285660444 @default.
• W3000207760 cites W2286365479 @default.
• W3000207760 cites W2289252105 @default.
• W3000207760 cites W2290132443 @default.
• W3000207760 cites W2294282016 @default.
• W3000207760 cites W2295652899 @default.
• W3000207760 cites W2300242332 @default.
• W3000207760 cites W2395579298 @default.
• W3000207760 cites W2508602506 @default.
• W3000207760 cites W2510516734 @default.
• W3000207760 cites W2515385951 @default.
• W3000207760 cites W2516141709 @default.
• W3000207760 cites W2520083297 @default.
• W3000207760 cites W25321933 @default.
• W3000207760 cites W2551814622 @default.
• W3000207760 cites W2562773490 @default.
• W3000207760 cites W2563587242 @default.
• W3000207760 cites W2565851976 @default.
• W3000207760 cites W2574797063 @default.
• W3000207760 cites W2584311934 @default.
• W3000207760 cites W2584616277 @default.
• W3000207760 cites W2585560244 @default.
• W3000207760 cites W2585774018 @default.
• W3000207760 cites W2592389822 @default.
• W3000207760 cites W2593564159 @default.
• W3000207760 cites W2594836184 @default.
• W3000207760 cites W2594928698 @default.
• W3000207760 cites W2605347906 @default.
• W3000207760 cites W2619096655 @default.
• W3000207760 cites W2625457103 @default.
• W3000207760 cites W2626991402 @default.
• W3000207760 cites W2657126969 @default.
• W3000207760 cites W2707890836 @default.
• W3000207760 cites W2739789140 @default.
• W3000207760 cites W2748818695 @default.
• W3000207760 cites W2788838111 @default.
• W3000207760 cites W2789246071 @default.
• W3000207760 cites W2790512080 @default.
• W3000207760 cites W2790925711 @default.
• W3000207760 cites W2792503273 @default.
• W3000207760 cites W2794470368 @default.
• W3000207760 cites W2794478957 @default.
• W3000207760 cites W2798170643 @default.
• W3000207760 cites W2801000640 @default.
• W3000207760 cites W2884687835 @default.
• W3000207760 cites W2886851211 @default.
• W3000207760 cites W2891561769 @default.
• W3000207760 cites W2894921321 @default.
• W3000207760 cites W2896942840 @default.
• W3000207760 cites W2900103278 @default.
• W3000207760 cites W2902332991 @default.
• W3000207760 cites W2904436509 @default.
• W3000207760 cites W2905104204 @default.
• W3000207760 cites W2915162395 @default.
• W3000207760 cites W2915803961 @default.
• W3000207760 cites W2916143956 @default.
• W3000207760 cites W2916321174 @default.
• W3000207760 cites W2916548364 @default.
• W3000207760 cites W2916892231 @default.
• W3000207760 cites W2916975147 @default.
• W3000207760 cites W2917307783 @default.
• W3000207760 cites W2917518248 @default.
• W3000207760 cites W2918006600 @default.
• W3000207760 cites W2930889563 @default.
• W3000207760 cites W2932154853 @default.
• W3000207760 cites W2950837708 @default.

• next