D-X-Y / Awesome-AutoDL

Automated Deep Learning: Neural Architecture Search Is Not the End (a curated list of AutoDL resources and an in-depth analysis)
MIT License
2.24k stars 314 forks source link

Please add NeurIPS 2019 paper #10

Closed negrinho closed 4 years ago

negrinho commented 4 years ago

Towards modular and programmable architecture search Paper: https://arxiv.org/abs/1909.13404 Code: https://github.com/negrinho/deep_architect

Thank you

D-X-Y commented 4 years ago

Nice paper, added it. Would you mind to let me know the type of its searching algorithm?

negrinho commented 4 years ago

The paper is more focused on designing a language to encode search spaces than on search algorithms. The language gives you a well-defined API with which search algorithms can interact with search spaces. This means that the implementations of search algorithms and search spaces can be decoupled to some extent. The code link I shared contains the implementation of the language and the framework that we built around it. In conclusion, this work is not about a new search algorithm or search space, but about how the introduction of this language can enable modularity in architecture search implementations. Hopefully that clarifies it :).

D-X-Y commented 4 years ago

@negrinho Very clear! Thanks for your explanation. Nice work.