Presentation 2023-06-30
On performance degradation of a method by minimizing the conditional mutual information for the out-of-distribution generalization
Genki Takahashi, Toshiyuki Tanaka,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In the out-of-distribution generalization problem, the smaller the degree of change in the data generating distribution during training and testing, the smaller the out-of-distribution generalization error of the model is expected to be. In this study, we devised a setting in which two distributions with an arbitrarily small KL divergence value can have an arbitrarily large difference in mutual information. This result suggests that out-of-distribution generalization methods based on conditional mutual information minimization can significantly increase the out-of-distribution generalization error even when the degree of change in the data generating distribution during training and testing is small.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Out-of-distribution generalization / Information theory / Representation learning
Paper # NC2023-15,IBISML2023-15
Date of Issue 2023-06-22 (NC, IBISML)

Conference Information
Committee NC / IBISML / IPSJ-BIO / IPSJ-MPS
Conference Date 2023/6/29(3days)
Place (in Japanese) (See Japanese page)
Place (in English) OIST Conference Center
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Hirokazu Tanaka(Tokyo City Univ.) / Masashi Sugiyama(Univ. of Tokyo)
Vice Chair Jun Izawa(Univ. of Tsukub) / Toshihiro Kamishima(AIST) / Koji Tsuda(Univ. of Tokyo)
Secretary Jun Izawa(NTT) / Toshihiro Kamishima(NAIST) / Koji Tsuda(NTT) / (Hokkaido Univ.)
Assistant Yoshimasa Tawatsuji(Waseda Univ.) / Takato Horii(Osaka Univ.) / Yoshinobu Kawahara(Osaka Univ.) / Taiji Suzuki(Tokyo Inst. of Tech.)

Paper Information
Registration To Technical Committee on Neurocomputing / Technical Committee on Information-Based Induction Sciences and Machine Learning / Special Interest Group on Bioinformatics and Genomics / Special Interest Group on Mathematical Modeling and Problem Solving
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) On performance degradation of a method by minimizing the conditional mutual information for the out-of-distribution generalization
Sub Title (in English)
Keyword(1) Out-of-distribution generalization
Keyword(2) Information theory
Keyword(3) Representation learning
1st Author's Name Genki Takahashi
1st Author's Affiliation Kyoto University(Kyoto University)
2nd Author's Name Toshiyuki Tanaka
2nd Author's Affiliation Kyoto University(Kyoto University)
Date 2023-06-30
Paper # NC2023-15,IBISML2023-15
Volume (vol) vol.123
Number (no) NC-90,IBISML-91
Page pp.pp.91-97(NC), pp.91-97(IBISML),
#Pages 7
Date of Issue 2023-06-22 (NC, IBISML)