Sampling hidden parameters from oracle distribution

Sho Sonoda, Noboru Murata

研究成果: Conference contribution

3 被引用数 (Scopus)

抄録

A new sampling learning method for neural networks is proposed. Derived from an integral representation of neural networks, an oracle probability distribution of hidden parameters is introduced. In general rigorous sampling from the oracle distribution holds numerical difficulty, a linear-time sampling algorithm is also developed. Numerical experiments showed that when hidden parameters were initialized by the oracle distribution, following backpropagation converged faster to better parameters than when parameters were initialized by a normal distribution.

本文言語English
ホスト出版物のタイトルArtificial Neural Networks and Machine Learning, ICANN 2014 - 24th International Conference on Artificial Neural Networks, Proceedings
出版社Springer Verlag
ページ539-546
ページ数8
ISBN(印刷版)9783319111780
DOI
出版ステータスPublished - 2014
イベント24th International Conference on Artificial Neural Networks, ICANN 2014 - Hamburg, Germany
継続期間: 2014 9 152014 9 19

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
8681 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

Other

Other24th International Conference on Artificial Neural Networks, ICANN 2014
国/地域Germany
CityHamburg
Period14/9/1514/9/19

ASJC Scopus subject areas

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Sampling hidden parameters from oracle distribution」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル