Awesome

DeepArchitect: Automatically Designing and Training Deep Architectures

IMPORTANT: This repo is not under active development. It contains a prototype for the ideas described in this paper. See our NeurIPS 2019 paper for the latest developments. The code and documentation for the latest framework can be found here.

This repository contains a Python implementation of the DeepArchitect framework described in our paper. To get familiar with the framework, we recommend starting with this tutorial.

A tar file with the logs of the experiments in the paper is available here. You can download it, unzip it in the top folder of the repo, and generate the plots of the paper using plots.py. The logs are composed of text and pickle files. It may be informative to inspect them. The experiments reported in the paper can be reproduced using experiments.py.

Contributors: Renato Negrinho, Geoff Gordon, Matt Gormley, Christoph Dann, Matt Barnes.

References

@article{negrinho2017deeparchitect,
  title={Deeparchitect: Automatically designing and training deep architectures},
  author={Negrinho, Renato and Gordon, Geoff},
  journal={arXiv preprint arXiv:1704.08792},
  year={2017}
}

@article{negrinho2019towards,
  title={Towards modular and programmable architecture search},
  author={Negrinho, Renato and Patil, Darshan and Le, Nghia and Ferreira, Daniel and Gormley, Matthew and Gordon, Geoffrey},
  journal={Neural Information Processing Systems},
  year={2019}
}