Summary of Generative Flow Induced Neural Architecture Search: Towards Discovering Optimal Architecture in Wavelet Neural Operator, by Hartej Soin and Tapas Tripura and Souvik Chakraborty
Generative flow induced neural architecture search: Towards discovering optimal architecture in wavelet neural operator
by Hartej Soin, Tapas Tripura, Souvik Chakraborty
First submitted to arxiv on: 11 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 generative flow-induced neural architecture search algorithm learns stochastic policies to generate sequences of architecture hyperparameters. This is achieved by using simple feed-forward neural networks that learn to generate states proportional to a reward from the terminal state. The algorithm demonstrates its efficacy on the wavelet neural operator (WNO), generating hyperparameters like wavelet basis and activation operators for wavelet integral blocks. The policy is learned by minimizing flow violation between states and maximizing reward from the terminal state, where WNO is trained simultaneously to guide the search. The proposed framework uses the exponent of the negative of WNO loss on the validation dataset as the reward function. Compared to grid search-based algorithms that foresee every combination, this method generates the most probable sequence based on positive reward, reducing exploration time. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to find good neural networks by generating sequences of architecture hyperparameters. This is done using simple computer programs that learn to make decisions based on rewards from the end result. The algorithm tests itself on wavelet neural operators, which are used for solving fluid mechanics problems. The method learns by finding patterns in successful architectures and then uses this knowledge to generate new architectures quickly. |
Keywords
» Artificial intelligence » Grid search
