Complex infomax: Convergence and approximation of infomax with complex nonlinearities

V. Calhoun, T. Adali

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Independent component analysis (ICA) for separating complex-valued sources is needed for convolutive source-separation in the frequency domain, or for performing source separation on complex-valued data, such as functional magnetic resonance imaging data. Previous complex infomax approaches have proposed using bounded (and hence non-analytic) nonlinearities. We propose using an analytic (and hence unbounded) complex nonlinearity for infomax for processing complex-valued sources. We show that using an analytic nonlinearity for processing complex data has a number of advantages. First, when compared to split-complex approaches (i.e., approaches that split the real and imaginary data into separate channels), the shape of the performance surface is improved resulting in better convergence characteristics. Additionally, the computational complexity is significantly reduced, and finally, the presence of cross terms in the Jacobian enables the analytic nonlinearity to approximate a more general class of input distributions.

Original languageEnglish (US)
Title of host publicationNeural Networks for Signal Processing XII - Proceedings of the 2002 IEEE Signal Processing Society Workshop, NNSP 2002
EditorsSamy Bengio, Scott Douglas, Tulay Adali, Jan Larsen, Herve Bourlard
PublisherInstitute of Electrical and Electronics Engineers Inc.
Number of pages10
ISBN (Electronic)0780376161
StatePublished - Jan 1 2002
Event12th IEEE Workshop on Neural Networks for Signal Processing, NNSP 2002 - Martigny, Switzerland
Duration: Sep 6 2002 → …

Publication series

NameNeural Networks for Signal Processing - Proceedings of the IEEE Workshop


Other12th IEEE Workshop on Neural Networks for Signal Processing, NNSP 2002
Period9/6/02 → …

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Artificial Intelligence
  • Software
  • Computer Networks and Communications
  • Signal Processing


Dive into the research topics of 'Complex infomax: Convergence and approximation of infomax with complex nonlinearities'. Together they form a unique fingerprint.

Cite this