Summary of Tg-nas: Leveraging Zero-cost Proxies with Transformer and Graph Convolution Networks For Efficient Neural Architecture Search, by Ye Qiao et al.
TG-NAS: Leveraging Zero-Cost Proxies with Transformer and Graph Convolution Networks for Efficient Neural Architecture Search
by Ye Qiao, Haocheng Xu, Sitao Huang
First submitted to arxiv on: 30 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 This paper introduces TG-NAS, a novel model-based universal proxy for neural architecture search (NAS). The goal is to create training-free proxies that predict architecture performance without requiring expensive evaluations. Existing approaches often rely on simple metrics like model parameter counts or floating-point operations, which are suboptimal and can’t be generalized to new search spaces. TG-NAS leverages a transformer-based operator embedding generator and a graph convolution network (GCN) to predict performance across any given search space. This proxy guides NAS without retraining, showing advantages in data independence, cost-effectiveness, and consistency. Experiments demonstrate its efficiency improvements over previous methods, discovering competitive models on CIFAR-10 and ImageNet. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary TG-NAS is a new way to find good designs for artificial intelligence (AI) models. Usually, we have to test many different designs to see which ones work best. But this can take a long time! TG-NAS lets us predict how well a design will do without actually testing it. This helps us find better designs faster and more easily. It works by using a special kind of computer program that understands the building blocks of AI models. With TG-NAS, we can search for good designs across many different kinds of problems without having to retrain our program each time. |
Keywords
* Artificial intelligence * Embedding * Gcn * Transformer
