Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Variational Dropout Sparsifies Deep Neural Networks #11

Open
ryoherisson opened this issue Jan 28, 2021 · 0 comments
Open

Variational Dropout Sparsifies Deep Neural Networks #11

ryoherisson opened this issue Jan 28, 2021 · 0 comments
Assignees

Comments

@ryoherisson
Copy link
Contributor

一言でいうと

変分ドロップアウトを,ドロップアウト率に束縛されないケースまで拡張し,勾配の分散を減らす方法を提案.
※変分ドロップアウト:各重み,レイヤーのドロップアウト率を個別にベイズ的に調整.

論文リンク

https://dl.acm.org/doi/pdf/10.5555/3305890.3305939

著者/所属機関

Dmitry Molchanov, Arsenii Ashukha, Dmitry Vetrov
(Yandex, Skolkovo Institute of Science and Technology, National Research University Higher
School of Economics, Moscow Institute of
Physics and Technology)

投稿日付(yyyy/MM/dd)

2017/06/13

概要

変分ドロップアウト率を0.0~1.0まで拡張し,勾配の分散を減らす手法を提案.
精度劣化を防ぎながら,パラメータ数を削減.

新規性・差分

  • 変分ドロップアウトを,すべてのドロップアウト率(0.0~1.0)まで拡張.
  • KL-Divergenceの新しい近似方法を提案.
  • 勾配の分散を減少させる方法を提案.

手法

全結合層のスパース変分ドロップアウト
スクリーンショット 2021-01-28 12 45 36

畳み込み層のスパース変分ドロップアウト
スクリーンショット 2021-01-28 12 45 45

結果

精度劣化を防ぎながら,LeNetでパラメータ数を最大280倍、VGGライクなネットワークで最大68倍まで削減.
スクリーンショット 2021-01-28 10 14 39

コメント

@ryoherisson ryoherisson self-assigned this Jan 28, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

2 participants