Upload
Download free for 30 days
Login
Submit Search
Squeeze and-excitation networks
0 likes
1,150 views
shima o
Summary of https://arxiv.org/abs/1709.01507
Science
Read more
1 of 8
Download now
Download to read offline
1
2
3
4
5
6
7
8
More Related Content
PPT
確率プロット
Tomoyuki Tarumi
PDF
Deeplearning4.4 takmin
Takuya Minagawa
PDF
バイアス付きMMSE-STSA推定器の理論解析およびミュージカルノイズフリー雑音抑圧への拡張
奈良先端大 情報科学研究科
PPTX
Graph Refinement based Tree Extraction using Mean-Field Networks and Graph Ne...
yukihiro domae
PDF
ICML2013読み会: Distributed training of Large-scale Logistic models
sleepy_yoshi
PPTX
ディープボルツマンマシン入門〜後半〜
sakaizawa
PDF
Chapter 8 ボルツマンマシン - 深層学習本読み会
Taikai Takeda
PDF
Prml5 6
K5_sem
確率プロット
Tomoyuki Tarumi
Deeplearning4.4 takmin
Takuya Minagawa
バイアス付きMMSE-STSA推定器の理論解析およびミュージカルノイズフリー雑音抑圧への拡張
奈良先端大 情報科学研究科
Graph Refinement based Tree Extraction using Mean-Field Networks and Graph Ne...
yukihiro domae
ICML2013読み会: Distributed training of Large-scale Logistic models
sleepy_yoshi
ディープボルツマンマシン入門〜後半〜
sakaizawa
Chapter 8 ボルツマンマシン - 深層学習本読み会
Taikai Takeda
Prml5 6
K5_sem
More from shima o
(20)
PDF
[読会]Causal transfer random forest combining logged data and randomized expe...
shima o
PDF
[読会]P qk means-_ billion-scale clustering for product-quantized codes
shima o
PDF
[読会]Long tail learning via logit adjustment
shima o
PDF
[読会]A critical review of lasso and its derivatives for variable selection und...
shima o
PDF
[読会]Themis decentralized and trustless ad platform with reporting integrity
shima o
PDF
[読会]Logistic regression models for aggregated data
shima o
PDF
Introduction of introduction_to_group_theory
shima o
PDF
Dynamic filters in graph convolutional network
shima o
PDF
Nmp for quantum_chemistry
shima o
PDF
Dl study g_learning_to_remember_rare_events
shima o
PDF
連続最適化勉強会
shima o
PDF
ReviewNet_161122
shima o
PDF
finite time analysis of the multiarmed bandit problem
shima o
PDF
normalized online learning
shima o
PDF
logistic regression in rare events data
shima o
PDF
Joint optimization of bid and budget allocation in sponsored search
shima o
PDF
Towards a robust modeling of temporal interest change patterns for behavioral...
shima o
PDF
Real time bidding algorithms for performance-based display ad allocation
shima o
PDF
Fingind the right consumer - optimizing for conversion in display advertising...
shima o
PDF
Real time bid optimization with smooth budget delivery in online advertising
shima o
[読会]Causal transfer random forest combining logged data and randomized expe...
shima o
[読会]P qk means-_ billion-scale clustering for product-quantized codes
shima o
[読会]Long tail learning via logit adjustment
shima o
[読会]A critical review of lasso and its derivatives for variable selection und...
shima o
[読会]Themis decentralized and trustless ad platform with reporting integrity
shima o
[読会]Logistic regression models for aggregated data
shima o
Introduction of introduction_to_group_theory
shima o
Dynamic filters in graph convolutional network
shima o
Nmp for quantum_chemistry
shima o
Dl study g_learning_to_remember_rare_events
shima o
連続最適化勉強会
shima o
ReviewNet_161122
shima o
finite time analysis of the multiarmed bandit problem
shima o
normalized online learning
shima o
logistic regression in rare events data
shima o
Joint optimization of bid and budget allocation in sponsored search
shima o
Towards a robust modeling of temporal interest change patterns for behavioral...
shima o
Real time bidding algorithms for performance-based display ad allocation
shima o
Fingind the right consumer - optimizing for conversion in display advertising...
shima o
Real time bid optimization with smooth budget delivery in online advertising
shima o
Ad
Squeeze and-excitation networks
1.
Squeeze‐and‐Excitation Networks, Hu+, '17
2.
Agenda 概要 手法 実験/結果
3.
概要 チャネル間の相互依存性を明示的に加えることでネットワークの表現力を上げる構造の 提案 Squeeze、Excitationの2つの機構を追加 既存の手法に追加するだけでよい パラメタ数の増加は微量 ImageNetでSoTA
4.
手法 Squeeze‐and‐Excitation Block(see Fig2,3) F
: transformation=Conv.層 F : sqeeze=Global average poolingで1×1×Cへ F : Excitation=bottleneck構造(パラメタrでくびれ率を決める) ここでチャネル間の関係性を考慮したweightを算出 : Uの各チャネルにF で算出したスカラ値s をかける tr sq ex X ~ ex c
5.
Squeeze‐and‐Excitation Block
6.
Squeeze‐and‐Excitation Block
7.
実験結果1 SE構造を導入しても処理速度はほぼ変わらない
8.
実験結果2
Download