FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4210486199> ?p ?o ?g. }

• W4210486199 endingPage "26" @default.
• W4210486199 startingPage "1" @default.
• W4210486199 abstract "Ternary Neural Networks (TNNs) and mixed-precision Ternary Binary Networks (TBNs) have demonstrated higher accuracy compared to Binary Neural Networks (BNNs) while providing fast, low-power, and memory-efficient inference. Related works have improved the accuracy of TNNs and TBNs, but overlooked their optimizations on CPU and GPU platforms. First, there is no unified encoding for the binary and ternary values in TNNs and TBNs. Second, existing works store the 2-bit quantized data sequentially in 32/64-bit integers, resulting in bit-extraction overhead. Last, adopting standard 2-bit multiplications for ternary values leads to a complex computation pipeline, and efficient mixed-precision multiplication between ternary and binary values is unavailable. In this article, we propose TAB as a unified and optimized inference method for ternary, binary, and mixed-precision neural networks. TAB includes unified value representation, efficient data storage scheme and novel bitwise dot product pipelines on CPU/GPU platforms. We adopt signed integers for consistent value representation across binary and ternary values. We introduce a bitwidth-last data format that stores the first and second bits of the ternary values separately to remove the bit extraction overhead. We design the ternary and binary bitwise dot product pipelines based on Gated-XOR using up to 40% fewer operations than State-Of-The-Art (SOTA) methods. Theoretical speedup analysis shows that our proposed TAB-TNN is 2.3× fast as the SOTA ternary method RTN, 9.8× fast as 8-bit integer quantization (INT8), and 39.4× fast as 32-bit full-precision convolution (FP32). Experiment results on CPU and GPU platforms show that our TAB-TNN has achieved up to 34.6× speedup and 16× storage size reduction compared with FP32 layers. TBN, Binary-activation Ternary-weight Network (BTN), and BNN in TAB are up to 40.7×, 56.2×, and 72.2× as fast as FP32. TAB-TNN is up to 70.1% faster and 12.8% more power-efficient than RTN on Darknet-19 while keeping the same accuracy. TAB is open source as a PyTorch Extension 1 for easy integration with existing CNN models." @default.
• W4210486199 created "2022-02-08" @default.
• W4210486199 creator A5016725699 @default.
• W4210486199 creator A5038598144 @default.
• W4210486199 creator A5087479741 @default.
• W4210486199 date "2022-09-30" @default.
• W4210486199 modified "2023-10-10" @default.
• W4210486199 title "TAB: Unified and Optimized Ternary, Binary, and Mixed-precision Neural Network Inference on the Edge" @default.
• W4210486199 cites W1863336885 @default.
• W4210486199 cites W2079658918 @default.
• W4210486199 cites W2300242332 @default.
• W4210486199 cites W2499931820 @default.
• W4210486199 cites W2618939455 @default.
• W4210486199 cites W2884150179 @default.
• W4210486199 cites W2887447938 @default.
• W4210486199 cites W2896934366 @default.
• W4210486199 cites W2954048742 @default.
• W4210486199 cites W2962939807 @default.
• W4210486199 cites W2963480671 @default.
• W4210486199 cites W2963526839 @default.
• W4210486199 cites W2982165597 @default.
• W4210486199 cites W2997394738 @default.
• W4210486199 cites W3008591352 @default.
• W4210486199 cites W3026337110 @default.
• W4210486199 cites W3034297393 @default.
• W4210486199 cites W3034933748 @default.
• W4210486199 cites W3034971973 @default.
• W4210486199 cites W3035232708 @default.
• W4210486199 cites W3128120039 @default.
• W4210486199 cites W3135547017 @default.
• W4210486199 cites W3151873305 @default.
• W4210486199 cites W3173877717 @default.
• W4210486199 cites W3174071941 @default.
• W4210486199 cites W3184606595 @default.
• W4210486199 cites W3211127556 @default.
• W4210486199 doi "https://doi.org/10.1145/3508390" @default.
• W4210486199 hasPublicationYear "2022" @default.
• W4210486199 type Work @default.
• W4210486199 citedByCount "3" @default.
• W4210486199 countsByYear W42104861992022 @default.
• W4210486199 countsByYear W42104861992023 @default.
• W4210486199 crossrefType "journal-article" @default.
• W4210486199 hasAuthorship W4210486199A5016725699 @default.
• W4210486199 hasAuthorship W4210486199A5038598144 @default.
• W4210486199 hasAuthorship W4210486199A5087479741 @default.
• W4210486199 hasBestOaLocation W42104861991 @default.
• W4210486199 hasConcept C11413529 @default.
• W4210486199 hasConcept C134765980 @default.
• W4210486199 hasConcept C154945302 @default.
• W4210486199 hasConcept C173608175 @default.
• W4210486199 hasConcept C199360897 @default.
• W4210486199 hasConcept C33923547 @default.
• W4210486199 hasConcept C41008148 @default.
• W4210486199 hasConcept C48372109 @default.
• W4210486199 hasConcept C50644808 @default.
• W4210486199 hasConcept C64452783 @default.
• W4210486199 hasConcept C94375191 @default.
• W4210486199 hasConceptScore W4210486199C11413529 @default.
• W4210486199 hasConceptScore W4210486199C134765980 @default.
• W4210486199 hasConceptScore W4210486199C154945302 @default.
• W4210486199 hasConceptScore W4210486199C173608175 @default.
• W4210486199 hasConceptScore W4210486199C199360897 @default.
• W4210486199 hasConceptScore W4210486199C33923547 @default.
• W4210486199 hasConceptScore W4210486199C41008148 @default.
• W4210486199 hasConceptScore W4210486199C48372109 @default.
• W4210486199 hasConceptScore W4210486199C50644808 @default.
• W4210486199 hasConceptScore W4210486199C64452783 @default.
• W4210486199 hasConceptScore W4210486199C94375191 @default.
• W4210486199 hasFunder F4320320751 @default.
• W4210486199 hasFunder F4320320766 @default.
• W4210486199 hasIssue "5" @default.
• W4210486199 hasLocation W42104861991 @default.
• W4210486199 hasLocation W42104861992 @default.
• W4210486199 hasOpenAccess W4210486199 @default.
• W4210486199 hasPrimaryLocation W42104861991 @default.
• W4210486199 hasRelatedWork W1509211761 @default.
• W4210486199 hasRelatedWork W1558545464 @default.
• W4210486199 hasRelatedWork W1642429638 @default.
• W4210486199 hasRelatedWork W1984303163 @default.
• W4210486199 hasRelatedWork W2117014006 @default.
• W4210486199 hasRelatedWork W2358725432 @default.
• W4210486199 hasRelatedWork W2372170743 @default.
• W4210486199 hasRelatedWork W2502560717 @default.
• W4210486199 hasRelatedWork W3047022145 @default.
• W4210486199 hasRelatedWork W4233815414 @default.
• W4210486199 hasVolume "21" @default.
• W4210486199 isParatext "false" @default.
• W4210486199 isRetracted "false" @default.
• W4210486199 workType "article" @default.