Kohei Hayashi / 林浩平

I'm a researcher working on machine learning at Preferred NetworksCurrently interested in the following topics:
  • Approximate Bayesian inference with variational methods and asymptotic statistics [AISTATS'18, ICML'15, ACML'15, NeurIPS'13, ICML'12]
  • Tensor and matrix decompositions [NeurIPS'17x2, ICML'17, IJCAI'17, ICML'16, AAAI'16, KDD'15, ECML'12, ICDM'10]
  • Web data mining [IJCAI'16, WSDM'16, KDD'15]
  • Generalization theory of deep neural nets [no paper yet]

Recent Publications

  • KH, Masaaki Imaizumi, Yuichi Yoshida
    On Random Subsampling of Gaussian Process Regression: A Graphon-Based Analysis
  • KH, Takuya Konishi, Tatsuro Kawamoto
    A Tractable Fully Bayesian Method for the Stochastic Block Model
    [arXiv, code]
  • Katsuhiko Ishiguro, KH
    Learning on Relational Data (in Japanese)
    Kodansha, 2016 [amazon]
Refereed journal, conference & workshop papers
  • Huda Hakami, KH, Danushka Bollegala
    Why does PairDiff work? - A Mathematical Analysis of Bilinear Relational Compositional Operators for Analogy Detection
    COLING, 2018 [arXiv]
  • Danushka Bollegala, KH, Ken-ichi Kawarabayashi
    Think Globally, Embed Locally — Locally Linear Meta-Embedding of Words
    IJCAI-ECAI, 2018 
    [arXiv, code]
  • Satoshi Hara, KH
    Making Tree Ensembles Interpretable: A Bayesian Model Selection Approach
    AISTATS, 2018 [arXivcode]
  • KH, Yuichi Yoshida
    Fitting Low-Rank Tensors in Constant Time
    NeurIPS2017 [pdf, videopostercode]
  • Masaaki Imaizumi, Takanori Maehara, KH
    On Tensor Train Rank Minimization: Statistical Efficiency and Scalable Algorithm
    NeurIPS, 2017 [pdfarXiv]
  • Danushka Bollegala, KH, Ken-ichi Kawarabayashi
    Learning Linear Transformations between Counting-based and Prediction-based Word Embeddings
    PLoS ONE 12(9): e0184544, 2017 [pdfdoirecipe]
  • Masaaki Imaizumi, KH
    Tensor Decomposition with Smoothness
    ICML, 2017 
  • Yuto Yamaguchi, KH
    Tensor Decomposition with Missing Indices 
    IJCAI, 2017 [pdf, slidescode]
  • Yuto Yamaguchi, KH
    When Does Label Propagation Fail? A View from a Network Generative Model 
    IJCAI, 2017 [pdf, slidescode]
  • Yohei Kondo, KH, Shin-ichi Maeda
    Sparse Bayesian linear regression with latent masking variables
    Neurocomputing, 2017 [doi]
  • KH, Yuichi Yoshida
    Minimizing Quadratic Functions in Constant Time 
    NeurIPS, 2016 [pdfarXiv, code]
  • Satoshi Hara, KH
    Making Tree Ensembles Interpretable

     ICML Workshop, 2016 [arXiv, code]
    Best paper runner-up. An extended version is now available.
  • Masaaki Imaizumi, KH
    Doubly Decomposing Nonparametric Tensor Regression 
    ICML, 2016 [pdf, arXiv, slides]
  • Takuya Konishi, Tomoharu Iwata, KH, Ken-ichi Kawarabayashi
    Identifying Key Observers to Find Popular Information in Advance
    IJCAI, 2016 [pdf]
  • Takanori Maehara, KHKen-ichi Kawarabayashi
    Expected Tensor Decomposition with Stochastic Gradient Descent
    AAAI, 2016 [pdf, slides(unofficial)]
  • Takuya Konishi, Takuya Ohwa, Sumio Fujita, Kazushi Ikeda, KH
    Extracting Search Query Patterns via the Pairwise Coupled Topic Model
    WSDM, 2016
  • Yohei Kondo, KH, Shin-ichi Maeda
    Bayesian Masking: Sparse Bayesian Estimation with Weaker Shrinkage Bias
    ACML, 2015 [pdfarXiv]
  • KH, Takanori Maehara, Masashi Toyoda, Ken-ichi Kawarabayashi
    Real-time Top-R Topic Detection on Twitter with Topic Hijack Filtering
    KDD, 2015 [pdfdoislides, code]
  • KH, Shin-ichi Maeda, Ryohei Fujimaki
    Rebuilding Factorized Information Criterion: Asymptotically Accurate Marginal Likelihood
    ICML, 2015 [pdf, slides]

Codes, etc

Work Experience

  • May 2018 -- Present
    Researcher, Preferred Networks
  • Jun 2016 -- Apr 2018
    Technical Adviser, LeapMind
  • Apr 2016 -- May 2018
    Researcher, National Institute of Advanced Industrial Science and Technology
  • Feb 2013 -- Mar 2016
    Project Researcher, National Institute of Informatics (Director: Prof. Ken-ichi Kawarabayashi)
  • Apr 2012 -- Jan 2013
    Postdoc, the University of Tokyo (Host: Prof. Hisashi Kashima)
  • Jan 2012 -- Mar 2012
    Intern, NEC (Mentor: Dr. Ryohei Fujimaki)


Mar 2012, D.Eng
  • Nara Institute of Science and Technology, Japan
  • Supervisor: Prof. Kazushi Ikeda.
  • Thesis: Generalization of tensor factorization and applications.
Mar 2009, M.Eng
  • Nara Institute of Science and Technology, Japan
  • Supervisor: Prof. Shin Ishii.
  • Thesis: Dynamic relational data analysis based on exponential family matrix factorization.
Mar 2007, B.Eng
  • Ritsumeikan University, Japan
  • Supervisor: Prof. Ikuko Nishikawa.
  • Thesis: CPG model for multiped locomotion by complex-valued neural networks.


  • PC: IBIS 2014AABI 2015, 2016.
  • Conference Reviewer: NeurIPS 2012, 2014--2018; ICML 2015, 2017--2019; ICLR 2018, 2019; AISTATS 2017--2019; ECMLPKDD 2013, 2015; WWW 2016; WSDM 2018; AAAI 2017--2018; IJCAI 2015, 2017; CIKM 2016.
  • Journal Reviewer: JMLR; IEEE TNNLS; DAMI; Neural Network; IEICE.


Kohei Hayashi,
Feb 28, 2013, 10:22 PM
Kohei Hayashi,
Dec 8, 2017, 11:53 AM