Matches in SemOpenAlex for { <https://semopenalex.org/work/W3094213821> ?p ?o ?g. }
- W3094213821 abstract "Multiplication (e.g., convolution) is arguably a cornerstone of modern deep neural networks (DNNs). However, intensive multiplications cause expensive resource costs that challenge DNNs' deployment on resource-constrained edge devices, driving several attempts for multiplication-less deep networks. This paper presented ShiftAddNet, whose main inspiration is drawn from a common practice in energy-efficient hardware implementation, that is, multiplication can be instead performed with additions and logical bit-shifts. We leverage this idea to explicitly parameterize deep networks in this way, yielding a new type of deep network that involves only bit-shift and additive weight layers. This hardware-inspired ShiftAddNet immediately leads to both energy-efficient inference and training, without compromising the expressive capacity compared to standard DNNs. The two complementary operation types (bit-shift and add) additionally enable finer-grained control of the model's learning capacity, leading to more flexible trade-off between accuracy and (training) efficiency, as well as improved robustness to quantization and pruning. We conduct extensive experiments and ablation studies, all backed up by our FPGA-based ShiftAddNet implementation and energy measurements. Compared to existing DNNs or other multiplication-less models, ShiftAddNet aggressively reduces over 80% hardware-quantified energy cost of DNNs training and inference, while offering comparable or better accuracies. Codes and pre-trained models are available at https://github.com/RICE-EIC/ShiftAddNet." @default.
- W3094213821 created "2020-10-29" @default.
- W3094213821 creator A5003718473 @default.
- W3094213821 creator A5013113668 @default.
- W3094213821 creator A5019582323 @default.
- W3094213821 creator A5048522863 @default.
- W3094213821 creator A5063291373 @default.
- W3094213821 creator A5065172226 @default.
- W3094213821 creator A5075140744 @default.
- W3094213821 creator A5080793133 @default.
- W3094213821 date "2020-10-24" @default.
- W3094213821 modified "2023-10-03" @default.
- W3094213821 title "ShiftAddNet: A Hardware-Inspired Deep Network" @default.
- W3094213821 cites W123295786 @default.
- W3094213821 cites W1659482141 @default.
- W3094213821 cites W1686810756 @default.
- W3094213821 cites W1693331953 @default.
- W3094213821 cites W1992348535 @default.
- W3094213821 cites W2004528524 @default.
- W3094213821 cites W2012557818 @default.
- W3094213821 cites W2120714355 @default.
- W3094213821 cites W2145343602 @default.
- W3094213821 cites W2194775991 @default.
- W3094213821 cites W2198190323 @default.
- W3094213821 cites W2319920447 @default.
- W3094213821 cites W2566876368 @default.
- W3094213821 cites W2612445135 @default.
- W3094213821 cites W2755682530 @default.
- W3094213821 cites W2759777444 @default.
- W3094213821 cites W2766224267 @default.
- W3094213821 cites W2890246415 @default.
- W3094213821 cites W2892104683 @default.
- W3094213821 cites W2896409484 @default.
- W3094213821 cites W2907718813 @default.
- W3094213821 cites W2946706208 @default.
- W3094213821 cites W2946958331 @default.
- W3094213821 cites W2948223045 @default.
- W3094213821 cites W2963551763 @default.
- W3094213821 cites W2963709416 @default.
- W3094213821 cites W2963813662 @default.
- W3094213821 cites W2963839617 @default.
- W3094213821 cites W2963844898 @default.
- W3094213821 cites W2970328156 @default.
- W3094213821 cites W2982619380 @default.
- W3094213821 cites W2990765879 @default.
- W3094213821 cites W2995197005 @default.
- W3094213821 cites W3015548926 @default.
- W3094213821 cites W3019205968 @default.
- W3094213821 cites W3023151209 @default.
- W3094213821 cites W3034344052 @default.
- W3094213821 cites W3035204914 @default.
- W3094213821 cites W3035414587 @default.
- W3094213821 cites W3039678359 @default.
- W3094213821 cites W3042493405 @default.
- W3094213821 cites W3087455782 @default.
- W3094213821 cites W3088841956 @default.
- W3094213821 cites W3097691692 @default.
- W3094213821 cites W3103732786 @default.
- W3094213821 cites W3120016950 @default.
- W3094213821 cites W3181458334 @default.
- W3094213821 doi "https://doi.org/10.48550/arxiv.2010.12785" @default.
- W3094213821 hasPublicationYear "2020" @default.
- W3094213821 type Work @default.
- W3094213821 sameAs 3094213821 @default.
- W3094213821 citedByCount "7" @default.
- W3094213821 countsByYear W30942138212021 @default.
- W3094213821 countsByYear W30942138212023 @default.
- W3094213821 crossrefType "posted-content" @default.
- W3094213821 hasAuthorship W3094213821A5003718473 @default.
- W3094213821 hasAuthorship W3094213821A5013113668 @default.
- W3094213821 hasAuthorship W3094213821A5019582323 @default.
- W3094213821 hasAuthorship W3094213821A5048522863 @default.
- W3094213821 hasAuthorship W3094213821A5063291373 @default.
- W3094213821 hasAuthorship W3094213821A5065172226 @default.
- W3094213821 hasAuthorship W3094213821A5075140744 @default.
- W3094213821 hasAuthorship W3094213821A5080793133 @default.
- W3094213821 hasBestOaLocation W30942138211 @default.
- W3094213821 hasConcept C104317684 @default.
- W3094213821 hasConcept C108583219 @default.
- W3094213821 hasConcept C111919701 @default.
- W3094213821 hasConcept C113775141 @default.
- W3094213821 hasConcept C114614502 @default.
- W3094213821 hasConcept C118524514 @default.
- W3094213821 hasConcept C119599485 @default.
- W3094213821 hasConcept C127413603 @default.
- W3094213821 hasConcept C138236772 @default.
- W3094213821 hasConcept C153083717 @default.
- W3094213821 hasConcept C154945302 @default.
- W3094213821 hasConcept C185592680 @default.
- W3094213821 hasConcept C2742236 @default.
- W3094213821 hasConcept C2776214188 @default.
- W3094213821 hasConcept C2780595030 @default.
- W3094213821 hasConcept C2984842247 @default.
- W3094213821 hasConcept C33923547 @default.
- W3094213821 hasConcept C41008148 @default.
- W3094213821 hasConcept C42935608 @default.
- W3094213821 hasConcept C55493867 @default.
- W3094213821 hasConcept C63479239 @default.
- W3094213821 hasConcept C79974875 @default.
- W3094213821 hasConcept C9390403 @default.