Summary of Dcp: Learning Accelerator Dataflow For Neural Network Via Propagation, by Peng Xu et al.
DCP: Learning Accelerator Dataflow for Neural Network via Propagation
by Peng Xu, Wenqi Shao, Mingyu Ding, Ping Luo
First submitted to arxiv on: 9 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Image and Video Processing (eess.IV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach called Dataflow Code Propagation (DCP) is proposed to automatically find the optimal dataflow for deep neural network (DNN) layers without requiring human effort. This efficient data-centric method translates hardware (HW) dataflow configurations into a code representation in a unified dataflow coding space, which can be optimized by backpropagating gradients given a DNN layer or network. DCP learns a neural predictor to efficiently update the dataflow codes towards the desired gradient directions to minimize various optimization objectives such as latency and energy. This approach can be easily generalized to unseen HW configurations in a zero-shot or few-shot learning manner, outperforming methods like GAMMA that require thousands of samples. Extensive experiments on representative models like MobileNet, ResNet, and ViT demonstrate the superiority of DCP. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary DCP is a new way to make deep neural networks run faster and use less energy. It does this by automatically finding the best way to move data around inside a computer chip. This means that computers can learn and process information much more quickly without needing special programming. The approach uses a type of artificial intelligence called backpropagation to find the best solution. This means that it can be used on different types of computer chips and even ones that haven’t been invented yet. It’s like having a super smart helper that can make computers work better without needing human involvement. |
Keywords
» Artificial intelligence » Backpropagation » Few shot » Neural network » Optimization » Resnet » Vit » Zero shot
