Summary of Ce-nas: An End-to-end Carbon-efficient Neural Architecture Search Framework, by Yiyang Zhao et al.
CE-NAS: An End-to-End Carbon-Efficient Neural Architecture Search Framework
by Yiyang Zhao, Yunzhuo Liu, Bo Jiang, Tian Guo
First submitted to arxiv on: 3 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Signal Processing (eess.SP)
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 proposed framework, CE-NAS, addresses the high carbon cost associated with neural architecture search (NAS) by dynamically adjusting GPU resources based on carbon intensity and balancing energy-efficient sampling and evaluation tasks. This approach leverages a reinforcement-learning agent, time-series transformer, and multi-objective optimizer to reduce the NAS search space and lower carbon emissions. CE-NAS achieves state-of-the-art results for both NAS datasets and open-domain NAS tasks while reducing carbon consumption by up to 7.22X. For example, on CIFAR-10, it achieves 97.35% top-1 accuracy with only 1.68M parameters and a carbon consumption of 38.53 lbs of CO2. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper presents a new way to design neural networks that uses less energy. This is important because designing neural networks can use a lot of energy, which contributes to climate change. The new approach, called CE-NAS, adjusts how it uses computer resources based on the amount of energy needed. It also tries to find good designs quickly and efficiently. The paper shows that this approach works well and reduces the energy used in designing neural networks by up to 7 times. |
Keywords
» Artificial intelligence » Reinforcement learning » Time series » Transformer
