SPLICE: Fully tractable hierarchical extension of ICA with pooling

Junichiro Hirayama, Aapo Johannes Hyvärinen, Motoaki Kawanabe

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review


We present a novel probabilistic framework for a hierarchical extension of independent component analysis (ICA), with a particular motivation in neuroscientific data analysis and modeling. The framework incorporates a general sub-space pooling with linear ICA-like layers stacked recursively. Unlike related previous models, our generative model is fully tractable: both the likelihood and the posterior estimates of latent variables can readily be computed with analytically simple formulae. The model is particularly simple in the case of complex-valued data since the pooling can be reduced to taking the modulus of complex numbers. Experiments on elec-troencephalography (EEG) and natural images demonstrate the validity of the method. Copyright 2017 by the author(s).
Original languageEnglish
Title of host publicationProceedings of the 34 th International Conference on Machine Learning, Sydney, Australia
EditorsDoina Precup, Yee Whye Teh
Number of pages12
PublisherInternational Machine Learning Society (IMLS)
Publication date2017
ISBN (Electronic)9781510855144
Publication statusPublished - 2017
MoE publication typeA4 Article in conference proceedings
EventInternational Conference on Machine Learning - Sydney, Australia
Duration: 6 Aug 201711 Aug 2017
Conference number: 34

Publication series

NameProceedings of Machine Learning Research
ISSN (Electronic)2640-3498

Fields of Science

  • Artificial intelligence
  • Learning systems
  • Neurophysiology, Analysis and modeling
  • Complex-valued
  • Generative model
  • Independent component analyses (ICA)
  • Latent variable
  • Modulus of complex numbers
  • Natural images
  • Probabilistic framework, Independent component analysis
  • 113 Computer and information sciences

Cite this