Matches in SemOpenAlex for { <https://semopenalex.org/work/W3128497428> ?p ?o ?g. }
- W3128497428 abstract "Neural Architecture Search (NAS), together with model scaling, has shown remarkable progress in designing high accuracy and fast convolutional architecture families. However, as neither NAS nor model scaling considers sufficient hardware architecture details, they do not take full advantage of the emerging datacenter (DC) accelerators. In this paper, we search for fast and accurate CNN model families for efficient inference on DC accelerators. We first analyze DC accelerators and find that existing CNNs suffer from insufficient operational intensity, parallelism, and execution efficiency. These insights let us create a DC-accelerator-optimized search space, with space-to-depth, space-to-batch, hybrid fused convolution structures with vanilla and depthwise convolutions, and block-wise activation functions. On top of our DC accelerator optimized neural architecture search space, we further propose a latency-aware compound scaling (LACS), the first multi-objective compound scaling method optimizing both accuracy and latency. Our LACS discovers that network depth should grow much faster than image size and network width, which is quite different from previous compound scaling results. With the new search space and LACS, our search and scaling on datacenter accelerators results in a new model series named EfficientNet-X. EfficientNet-X is up to more than 2X faster than EfficientNet (a model series with state-of-the-art trade-off on FLOPs and accuracy) on TPUv3 and GPUv100, with comparable accuracy. EfficientNet-X is also up to 7X faster than recent RegNet and ResNeSt on TPUv3 and GPUv100." @default.
- W3128497428 created "2021-02-15" @default.
- W3128497428 creator A5002729731 @default.
- W3128497428 creator A5026512168 @default.
- W3128497428 creator A5038861110 @default.
- W3128497428 creator A5050207942 @default.
- W3128497428 creator A5064543333 @default.
- W3128497428 creator A5084000289 @default.
- W3128497428 creator A5088551093 @default.
- W3128497428 date "2021-02-10" @default.
- W3128497428 modified "2023-09-26" @default.
- W3128497428 title "Searching for Fast Model Families on Datacenter Accelerators" @default.
- W3128497428 cites W2002555321 @default.
- W3128497428 cites W2117539524 @default.
- W3128497428 cites W2183341477 @default.
- W3128497428 cites W2194775991 @default.
- W3128497428 cites W2274287116 @default.
- W3128497428 cites W2279098554 @default.
- W3128497428 cites W2482315298 @default.
- W3128497428 cites W2556833785 @default.
- W3128497428 cites W2592875630 @default.
- W3128497428 cites W2606722458 @default.
- W3128497428 cites W2612445135 @default.
- W3128497428 cites W2736043560 @default.
- W3128497428 cites W2752782242 @default.
- W3128497428 cites W2767286248 @default.
- W3128497428 cites W2771727678 @default.
- W3128497428 cites W2785366763 @default.
- W3128497428 cites W2788853733 @default.
- W3128497428 cites W2798724095 @default.
- W3128497428 cites W2810271953 @default.
- W3128497428 cites W2886851211 @default.
- W3128497428 cites W2897275599 @default.
- W3128497428 cites W2921495890 @default.
- W3128497428 cites W2951104886 @default.
- W3128497428 cites W2952975409 @default.
- W3128497428 cites W2954234207 @default.
- W3128497428 cites W2955425717 @default.
- W3128497428 cites W2960010704 @default.
- W3128497428 cites W2962697884 @default.
- W3128497428 cites W2962746461 @default.
- W3128497428 cites W2962782102 @default.
- W3128497428 cites W2963000224 @default.
- W3128497428 cites W2963163009 @default.
- W3128497428 cites W2963374479 @default.
- W3128497428 cites W2963674932 @default.
- W3128497428 cites W2963918968 @default.
- W3128497428 cites W2964259004 @default.
- W3128497428 cites W2964331719 @default.
- W3128497428 cites W2967733054 @default.
- W3128497428 cites W2970971581 @default.
- W3128497428 cites W2982083293 @default.
- W3128497428 cites W2985232804 @default.
- W3128497428 cites W2987515519 @default.
- W3128497428 cites W2994749257 @default.
- W3128497428 cites W3016163932 @default.
- W3128497428 cites W3016265891 @default.
- W3128497428 cites W3035130950 @default.
- W3128497428 cites W3049519473 @default.
- W3128497428 cites W2809297101 @default.
- W3128497428 doi "https://doi.org/10.48550/arxiv.2102.05610" @default.
- W3128497428 hasPublicationYear "2021" @default.
- W3128497428 type Work @default.
- W3128497428 sameAs 3128497428 @default.
- W3128497428 citedByCount "0" @default.
- W3128497428 crossrefType "posted-content" @default.
- W3128497428 hasAuthorship W3128497428A5002729731 @default.
- W3128497428 hasAuthorship W3128497428A5026512168 @default.
- W3128497428 hasAuthorship W3128497428A5038861110 @default.
- W3128497428 hasAuthorship W3128497428A5050207942 @default.
- W3128497428 hasAuthorship W3128497428A5064543333 @default.
- W3128497428 hasAuthorship W3128497428A5084000289 @default.
- W3128497428 hasAuthorship W3128497428A5088551093 @default.
- W3128497428 hasBestOaLocation W31284974281 @default.
- W3128497428 hasConcept C111919701 @default.
- W3128497428 hasConcept C113775141 @default.
- W3128497428 hasConcept C11413529 @default.
- W3128497428 hasConcept C123657996 @default.
- W3128497428 hasConcept C142362112 @default.
- W3128497428 hasConcept C153349607 @default.
- W3128497428 hasConcept C154945302 @default.
- W3128497428 hasConcept C173608175 @default.
- W3128497428 hasConcept C2524010 @default.
- W3128497428 hasConcept C2776214188 @default.
- W3128497428 hasConcept C2777210771 @default.
- W3128497428 hasConcept C2778572836 @default.
- W3128497428 hasConcept C33923547 @default.
- W3128497428 hasConcept C3826847 @default.
- W3128497428 hasConcept C41008148 @default.
- W3128497428 hasConcept C45347329 @default.
- W3128497428 hasConcept C459310 @default.
- W3128497428 hasConcept C50644808 @default.
- W3128497428 hasConcept C76155785 @default.
- W3128497428 hasConcept C81363708 @default.
- W3128497428 hasConcept C82876162 @default.
- W3128497428 hasConcept C99844830 @default.
- W3128497428 hasConceptScore W3128497428C111919701 @default.
- W3128497428 hasConceptScore W3128497428C113775141 @default.
- W3128497428 hasConceptScore W3128497428C11413529 @default.
- W3128497428 hasConceptScore W3128497428C123657996 @default.