Summary of The Devil Is in Discretization Discrepancy. Robustifying Differentiable Nas with Single-stage Searching Protocol, by Konstanty Subbotko et al.
The devil is in discretization discrepancy. Robustifying Differentiable NAS with Single-Stage Searching Protocol
by Konstanty Subbotko, Wojciech Jablonski, Piotr Bilinski
First submitted to arxiv on: 26 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); 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 The abstract discusses Neural Architecture Search (NAS) for computer vision tasks, specifically differentiable NAS (DNAS), where the optimal architecture is found through gradient-based methods. However, these methods suffer from discretization error, which can severely impact the process of obtaining the final architecture. The authors introduce a novel single-stage searching protocol that outperforms other DNAS methods by achieving 75.3% on the Cityscapes validation dataset and attaining performance 1.1% higher than DCNAS on non-dense search spaces. This approach is computationally efficient, with an entire training process taking only 5.5 GPU days due to weight reuse. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to design artificial intelligence models that can do tasks like image recognition and object detection. The current method of designing these models has some problems, like not being very accurate or taking too long to train. To fix this, the authors developed a new approach that is faster and more efficient. They tested their approach on a dataset called Cityscapes and found that it was able to achieve better results than other methods. This could be useful for things like self-driving cars or medical image analysis. |
Keywords
» Artificial intelligence » Object detection
