Presentation 2017-03-06
New Lerning Algorythm of Neural Network using Integral Representation and Kernel Herding
Takuo Matsubara, Sho Sonoda, Noboru Murata,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) A new learning algorithm for neural networks that converges at $mathcal{O}(frac{1}{n})$ with respect to model complexity $n$ is proposed. First, we derive an RKHS (reproducing kernel Hilbert space) version of the integral representation of neural networks. Then, combined with kernel herding, we propose a new learning algorithm for neural networks. Numerical experiments supported the theoretical analysis that our method converges at $mathcal{O}(frac{1}{n})$, which is faster than known rate, $mathcal{O}(frac{1}{sqrt{n}})$.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Neural Network / Integral Representation / RKHS / Kernel Herding / Convergence rate
Paper # IBISML2016-103
Date of Issue 2017-02-27 (IBISML)

Conference Information
Committee IBISML
Conference Date 2017/3/6(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Tokyo Institute of Technology
Topics (in Japanese) (See Japanese page)
Topics (in English) Statistical Mathematics, Machine Learning, Data Mining, etc.
Chair Kenji Fukumizu(ISM)
Vice Chair Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Kyoto Univ.)
Secretary Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Nagoya Inst. of Tech.)
Assistant Toshihiro Kamishima(AIST) / Tomoharu Iwata(NTT)

Paper Information
Registration To Technical Committee on Infomation-Based Induction Sciences and Machine Learning
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) New Lerning Algorythm of Neural Network using Integral Representation and Kernel Herding
Sub Title (in English)
Keyword(1) Neural Network
Keyword(2) Integral Representation
Keyword(3) RKHS
Keyword(4) Kernel Herding
Keyword(5) Convergence rate
1st Author's Name Takuo Matsubara
1st Author's Affiliation Waseda University(Waseda Univ.)
2nd Author's Name Sho Sonoda
2nd Author's Affiliation Waseda University(Waseda Univ.)
3rd Author's Name Noboru Murata
3rd Author's Affiliation Waseda University(Waseda Univ.)
Date 2017-03-06
Paper # IBISML2016-103
Volume (vol) vol.116
Number (no) IBISML-500
Page pp.pp.25-31(IBISML),
#Pages 7
Date of Issue 2017-02-27 (IBISML)