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

Towards Compact CNNs via Collaborative Compression #31

Open
ryoherisson opened this issue Jun 9, 2021 · 0 comments
Open

Towards Compact CNNs via Collaborative Compression #31

ryoherisson opened this issue Jun 9, 2021 · 0 comments

Comments

@ryoherisson
Copy link
Contributor

ryoherisson commented Jun 9, 2021

一言でいうと

チャネルプルーニングとテンソル分解を組み合わせてCNNモデルを圧縮するCollaborative Compression(CC)方式を提案.

論文リンク

https://arxiv.org/pdf/2105.11228.pdf

著者/所属機関

Yuchao Li, Shaohui Lin, Jianzhuang Liu, Qixiang Ye, Mengdi Wang, Fei Chao, Fan Yang, Jincheng Ma, Qi Tian, Rongrong Ji
(Xiamen University, Alibaba Group, East China Normal University
Huawei Noah’s Ark Lab, University of Chinese Academy of Sciences, Huawei TechnologiesCo.,Ltd, Xiamen University, Peng Cheng Laboratory)

投稿日付(yyyy/MM/dd)

2021/05/24

概要

モデル圧縮手法であるチャネルプルーニングとテンソル分解は,従来,単独で展開されていたため,高い圧縮率を追求すると精度が著しく劣化するという課題があった.

本論文では,モデルのスパース性と低ランク性を同時に学習することで,チャネルプルーニングとテンソル分解を組み合わせてCNNモデルを圧縮する手法を提案.

ResNet-50では,48.4%のパラメータを削除することで52.9%のFLOPs削減を達成し,ImageNet 2012ではトップ1からの精度劣化を0.56%を抑えた.

新規性・差分

  • グローバル圧縮率最適化とマルチステップヒューリスティック圧縮によるチャネルプルーニングとテンソル分解の組み合わせにより精度劣化を防ぎながらCNNを圧縮する手法を提案.
  • 各層の圧縮率を自動で決定.

手法

2段階で圧縮.

  • グローバル圧縮率最適化:圧縮感度に基づいた最適化問題を解き,各層の圧縮率を決定.
  • マルチステップヒューリスティック圧縮:重要度の低い圧縮ユニットを段階的に取り除くことで,各層を個別に圧縮.

スクリーンショット 2021-06-10 8 44 55

グローバル圧縮率最適化
スクリーンショット 2021-06-10 9 40 17
スクリーンショット 2021-06-10 9 40 29

マルチステップヒューリスティック圧縮
圧縮ユニットの重要度
スクリーンショット 2021-06-10 9 43 16

結果

スクリーンショット 2021-06-10 9 26 46
スクリーンショット 2021-06-10 9 27 02

コメント

@ryoherisson ryoherisson self-assigned this Jun 9, 2021
@ryoherisson ryoherisson changed the title [WIP]Towards Compact CNNs via Collaborative Compression Towards Compact CNNs via Collaborative Compression Jun 10, 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

1 participant