Summary of Swap-nas: Sample-wise Activation Patterns For Ultra-fast Nas, by Yameng Peng et al.
SWAP-NAS: Sample-Wise Activation Patterns for Ultra-fast NAS
by Yameng Peng, Andy Song, Haytham M. Fayek, Vic Ciesielski, Xiaojun Chang
First submitted to arxiv on: 7 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Neural and Evolutionary Computing (cs.NE)
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 Training-free metrics are crucial in Neural Architecture Search (NAS) to avoid resource-intensive neural network training. However, existing methods have limitations, including limited correlation and poor generalization across different search spaces and tasks. To address these issues, we propose Sample-Wise Activation Patterns (SWAP) and its derivative, SWAP-Score, a novel high-performance training-free metric. SWAP-Score measures the expressivity of networks over a batch of input samples and shows strong correlations with ground-truth performance across various search spaces and tasks. Our results outperform 15 existing training-free metrics on NAS-Bench-101/201/301 and TransNAS-Bench-101. Furthermore, regularisation can enhance the SWAP-Score, leading to even higher correlations in cell-based search space and enabling model size control during the search. For example, the Spearman’s rank correlation coefficient between regularised SWAP-Score and CIFAR-100 validation accuracies on NAS-Bench-201 networks is 0.90. Our SWAP-NAS achieves competitive performance on CIFAR-10 and ImageNet in approximately 6 minutes and 9 minutes of GPU time respectively. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you have a way to measure how good a new neural network architecture is without having to train it from scratch. This would save lots of computer time and energy! Researchers have been trying to develop such methods, but they have some limitations. Our team proposes a new approach called Sample-Wise Activation Patterns (SWAP) that can accurately predict the performance of a network without training it. We tested our method on many different networks and tasks and found that it works really well. In fact, it outperforms other existing methods by a lot! This means we can use this method to quickly find the best neural network architecture for certain tasks, like image recognition or speech recognition. |
Keywords
* Artificial intelligence * Generalization * Neural network
