# SCARLET-NAS: Bridging the Gap between Stability and Scalability in Weight-sharing Neural Architecture Search

arXiv: https://arxiv.org/abs/1908.06022 · HTML: https://mlsanity.com/paper/1908.06022/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 16 Aug 2019.

## Built on by (most cited first, 6 of 6)

- [Reconsidering CO2 emissions from Computer Vision](https://mlsanity.com/paper/2104.08702/) (2021)
- [β-DARTS++: Bi-level Regularization for Proxy-robust Differentiable Architecture Search](https://mlsanity.com/paper/2301.06393/) (2023)
- [Joint Learning of Neural Transfer and Architecture Adaptation for Image Recognition](https://mlsanity.com/paper/2103.16889/) (2021)
- [Improving Ranking Correlation of Supernet with Candidates Enhancement and Progressive Training](https://mlsanity.com/paper/2108.05866/) (2021)
- [Revisiting Architecture-aware Knowledge Distillation: Smaller Models and Faster Search](https://mlsanity.com/paper/2206.13130/) (2022)
- [Exploring Complicated Search Spaces with Interleaving-Free Sampling](https://mlsanity.com/paper/2112.02488/) (2021)

## Abstract (arXiv)

To discover powerful yet compact models is an important goal of neural architecture search. Previous two-stage one-shot approaches are limited by search space with a fixed depth. It seems handy to include an additional skip connection in the search space to make depths variable. However, it creates a large range of perturbation during supernet training and it has difficulty giving a confident ranking for subnetworks. In this paper, we discover that skip connections bring about significant feature inconsistency compared with other operations, which potentially degrades the supernet performance. Based on this observation, we tackle the problem by imposing an equivariant learnable stabilizer to homogenize such disparities. Experiments show that our proposed stabilizer helps to improve the supernet's convergence as well as ranking performance. With an evolutionary search backend that incorporates the stabilized supernet as an evaluator, we derive a family of state-of-the-art architectures, the SCARLET series of several depths, especially SCARLET-A obtains 76.9% top-1 accuracy on ImageNet. Code is available at https://github.com/xiaomi-automl/ScarletNAS.