Presentation 2005/6/16
Statistical Mechanism of Mutual Learning with Latent Teacher
Kazuyuki HARA, Masato OKADA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In Mutual learning, two learning machines (students) learn each other and a student acts as a teacher to another student. Therefore, teacher is not used in mutual learning. We analyze dynamics of mutual learning with latent teacher by using a statistical mechanics method. We assume that overlap between latent teacher and students are not zero before the learning. From our analysis, we show that the overlap between the latent teacher and student becomes larger after mutual learning while students did not learn teacher. We also show that asymptotic property of the generalization error of mutual learning is equal to that of Bagging uses not learned students. At the early stage of mutual learning, generalization error becomes minimum, and the generalization error at minimum is larger than that of parallel boosting uses not learned students, are shown.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Mutual learning / Latent teacher / on-line learning / ensemble learning / Statistical mechanics
Paper # NC2005-11
Date of Issue

Conference Information
Committee NC
Conference Date 2005/6/16(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Statistical Mechanism of Mutual Learning with Latent Teacher
Sub Title (in English)
Keyword(1) Mutual learning
Keyword(2) Latent teacher
Keyword(3) on-line learning
Keyword(4) ensemble learning
Keyword(5) Statistical mechanics
1st Author's Name Kazuyuki HARA
1st Author's Affiliation Department of Electronics and Information, Tokyo Metropolitan College of Technology()
2nd Author's Name Masato OKADA
2nd Author's Affiliation Division, Osaka Corporation:Laboratory for Mathematical Neuroscience, Brain Science Institute, RIKEN:Intellegent Cooperation and Control, PRESTO, Japan Science and Technology Agency
Date 2005/6/16
Paper # NC2005-11
Volume (vol) vol.105
Number (no) 130
Page pp.pp.-
#Pages 6
Date of Issue