Summary of Algorithmic Strategies For Sustainable Reuse Of Neural Network Accelerators with Permanent Faults, by Youssef A. Ait Alama et al.
Algorithmic Strategies for Sustainable Reuse of Neural Network Accelerators with Permanent Faults
by Youssef A. Ait Alama, Sampada Sakpal, Ke Wang, Razvan Bunescu, Avinash Karanth, Ahmed Louri
First submitted to arxiv on: 17 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Hardware Architecture (cs.AR)
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 algorithmic approach is proposed to mitigate permanent hardware faults in neural network (NN) accelerators based on systolic arrays. The existing solutions include localizing and isolating faulty processing elements, using redundant elements for re-execution, or decommissioning the accelerator. In this paper, a CUDA-accelerated systolic array simulator is introduced in PyTorch to quantify the impact of permanent faults on links connecting two PEs or in weight registers. The proposed techniques do not require hardware modification and rely on existing components of widely used accelerators. Extensive experimental evaluations using fully connected and convolutional NNs trained on MNIST, CIFAR-10, and ImageNet show that the proposed approach matches or gets close to original fault-free accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way is found to use faulty parts in neural network computers instead of throwing them away. When a part fails, it’s not used anymore, but this can waste resources. The researchers created a special computer program that simulates how systolic arrays work and tested their ideas on different types of neural networks. |
Keywords
» Artificial intelligence » Neural network
