Skip to content

A Computation of Visual Attention with Fisher Information Matrix

Notifications You must be signed in to change notification settings

yumaloop/AttFisher

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AttFisher: A Computation of Visual Attention with Fisher Information Matrix

はじめに

近年,機械学習分野では人の注意機構をヒントとしたAttentionと呼ばれるモデルアーキテクチャが注目されている. また,神経科学・生理学の分野では,特に視覚型注意に関する研究が盛んで, top-down attentionとbottom-up attentionが知られている.

  • Bottom-up attention: 目的に対して偏りに対する注意.
  • Top-down attention: 周りとの違いが顕著な刺激に対する注意.

Fisher情報量とKL-divergence

Fisher情報量

未知母数 に依存する 上の確率変数 の 対数尤度関数 に対して,Score関数

が定義される. ここで, の1次モーメントは となることが知られており, 2次モーメント をFisher情報量行列と呼び, これは, の対数尤度関数 に対するHessian行列の期待値として以下のように定義される.

KL-divergence

確率分布間の擬距離であるKL-divergence

について,2次までのMaclaurin展開をすると,

となる.

確率モデルとの関係.

...

Experiments

VAE on MNIST

img0

img0

img0

img0

About

A Computation of Visual Attention with Fisher Information Matrix

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published