fillassuncao/f-denser repository overview

⁠Fast-DENSER++ Docker Image

F-DENSER++ is a new extension to Deep Evolutionary Network Structured Evolution (DENSER). The vast majority of NeuroEvolution methods that optimise Deep Artificial Neural Networks (DANNs) only evaluate the candidate solutions for a fixed amount of epochs; this makes it difficult to effectively assess the learning strategy, and requires the best generated network to be further trained after evolution. F-DENSER++ enables the training time of the candidate solutions to grow continuously as necessary, i.e., in the initial generations the candidate solutions are trained for shorter times, and as generations proceed it is expected that longer training cycles enable better performances. Consequently, the models discovered by F-DENSER++ are fully-trained DANNs, and are ready for deployment after evolution, without the need for further training.

Details about the method can be found in:

@article{assunccao2019fast,
  title={Fast-DENSER++: Evolving Fully-Trained Deep Artificial Neural Networks},
  author={Assun{\c{c}}{\~a}o, Filipe and Louren{\c{c}}o, Nuno and Machado, Penousal and Ribeiro, Bernardete},
  journal={arXiv preprint arXiv:1905.02969},
  year={2019}
}

@article{assunccao2018denser,
  title={DENSER: deep evolutionary network structured representation},
  author={Assun{\c{c}}ao, Filipe and Louren{\c{c}}o, Nuno and Machado, Penousal and Ribeiro, Bernardete},
  journal={Genetic Programming and Evolvable Machines},
  pages={1--31},
  year={2018},
  publisher={Springer}
}

⁠Image Tags

• fillassuncao/f-denser:cpu -- CPU based code
• fillassuncao/f-denser:gpu -- GPU based code

⁠Running containers

CPU:

docker run -it -v $PWD/f-denser-files:/f-denser/experiments -w /f-denser fillassuncao/f-denser:cpu python f_denser.py -c example/config.json -g example/cnn.grammar -d fashion-mnist

GPU:

docker run -it --runtime=nvidia -v $PWD/f-denser-files:/f-denser/experiments -w /f-denser fillassuncao/f-denser:gpu python f_denser.py -c example/config.json -g example/cnn.grammar -d fashion-mnist

The path where the generated files are saved can be changed by modifying "-v $PWD/f-denser-files:/f-denser/experiments", which takes the format "-v local/path:container/path".

⁠Github repository

The codebase for F-DENSER++ can be found in http://www.github.com/fillassuncao/f-denser⁠. It includes a full description of the input parameters.

⁠Support

Any questions, comments or suggestion should be directed to Filipe Assunção ([email protected]⁠)