DCUF

Sample code of Disentangling controllable and uncontrollable factors of variation (DCUF) by Interacting with the World

Dependencies

• python version 2.7.6
• chainer version 1.8.2

Usage

python main.py ${LAMBDA} ${ICF_EPOCH} ${AE_EPOCH} ${DCUF_EPOCH}

${LAMBDA}: hyparparameter to balance autoencoder (AE) and disentangled objectives
${ICF_EPOCH}: the number of epoch to pretrain independently controllable factor's model
${AE_EPOCH}: the number of epoch to pretrain second AE
${DCUF_EPOCH}: the number of epoch to train DCUF model

Citation

Y Sawada, L Rigazio, K Morikawa, M Iwasaki, Y Bengio,
"Disentangling Controllable and Uncontrollable Factors by Interacting with the World", Deep RL Workshop NeurIPS 2018 https://sites.google.com/view/deep-rl-workshop-nips-2018/home
https://drive.google.com/open?id=0B_utB5Y8Y6D5UWVUMkhSckRjZTdKdTk5ZWxxRXVNaWNtOVpB

License

Copyright (c) 2018 Yoshihide Sawada
Released under the MIT license
https://opensource.org/licenses/mit-license.php