KAPLA: Pragmatic Representation and Fast Solving of Scalable NN Accelerator Dataflow
Not on the map. Why?
Abstract (arXiv)
Dataflow scheduling decisions are of vital importance to neural network (NN) accelerators. Recent scalable NN accelerators support a rich set of advanced dataflow techniques. The problems of comprehensively representing and quickly finding optimized dataflow schemes thus become significantly more complicated and challenging. In this work, we first propose comprehensive and pragmatic dataflow representations for temporal and spatial scheduling on scalable multi-node NN architectures. An informal hierarchical taxonomy highlights the tight coupling across different levels of the dataflow space as the major difficulty for fast design exploration. A set of formal tensor-centric directives accurately express various inter-layer and intra-layer schemes, and allow for quickly determining their validity and efficiency. We then build a generic, optimized, and fast dataflow solver, KAPLA, which makes use of the pragmatic directives to explore the design space with effective validity check and efficiency estimation. KAPLA decouples the upper inter-layer level for fast pruning, and solves the lower intra-layer schemes with a novel bottom-up cost descending method. KAPLA achieves within only 2.2% and 7.7% energy overheads on the result dataflow for training and inference, respectively, compared to the exhaustively searched optimal schemes. It also outperforms random and machine-learning-based approaches, with more optimized results and orders of magnitude faster search speedup.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 26 of its 60 references that are in the map. First posted on arXiv 9 Jun 2023.
Builds on
26 of 60 references.
References · 26
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,364
- 2014Going Deeper with Convolutionscited by 4,241
- 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,845
- 2014Sequence to Sequence Learning with Neural Networkscited by 2,986
- 2016Neural Architecture Search with Reinforcement Learningcited by 1,499
- 2018MnasNet: Platform-Aware Neural Architecture Search for Mobilecited by 943
- 2018Efficient Neural Architecture Search via Parameter Sharingcited by 837
- 2019Large Batch Optimization for Deep Learning: Training BERT in 76 minutescited by 674
- 2017In-Datacenter Performance Analysis of a Tensor Processing Unitcited by 618
- 2018FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Searchcited by 485
- 2016EIE: Efficient Inference Engine on Compressed Deep Neural Networkcited by 295
14 more
- 2018Beyond Data and Model Parallelism for Deep Neural Networkscited by 179
- 2018Learning to Optimize Tensor Programscited by 140
- 2017SCNN: An Accelerator for Compressed-sparse Convolutional Neural Networkscited by 135
- 2017Device Placement Optimization with Reinforcement Learningcited by 86
- 2018Interstellar: Using Halide's Scheduling Language to Analyze DNN Acceleratorscited by 34
- 2016Maximizing CNN Accelerator Efficiency Through Resource Partitioningcited by 29
- 2018Optimizing CNN Model Inference on CPUscited by 29
- 2020Chameleon: Adaptive Code Optimization for Expedited Deep Neural Network Compilationcited by 29
- 2021Mind Mappings: Enabling Efficient Algorithm-Accelerator Mapping Space Searchcited by 24
- 2017Extremely Large Minibatch SGD: Training ResNet-50 on ImageNet in 15 Minutescited by 22
- 2021CoSA: Scheduling by Constrained Optimization for Spatial Acceleratorscited by 22
- 2019HyPar: Towards Hybrid Parallelism for Deep Learning Accelerator Arraycited by 15
- 2021TENET: A Framework for Modeling Tensor Dataflow Based on Relation-centric Notationcited by 9
- 2016A Systematic Approach to Blocking Convolutional Neural Networkscited by 7
Built on by
Not cited yet.