This Shakeout implementation is based on caffe.
First, you should download the caffe
source code.
Then put the files shakeout_layer.cpp
; shakeout_layer.cu
under the directory src/caffe/layers/
;
put the file caffe.proto
under the directory src/caffe/proto/
;
put the file shakeout_layer.hpp
under the directory include/caffe/layers/
.
Finally, you can refer to AlexNet_shakeout_example.prototxt
to see how to use.
Please cite our Shakeout
paper in your publications if it helps your research:
@inproceedings{kang2016shakeout,
title={Shakeout: A New Regularized Deep Neural Network Training Scheme},
author={Kang, Guoliang and Li, Jun and Tao, Dacheng},
booktitle={Thirtieth AAAI Conference on Artificial Intelligence},
year={2016}
}
@article{kang2017shakeout,
title={Shakeout: A New Approach to Regularized Deep Neural Network Training},
author={Kang, Guoliang and Li, Jun and Tao, Dacheng},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2017},
publisher={IEEE}
}