Infomax-ICA using Hessian-free optimization

Philippe Tillet; H. T. Kung; David D. Cox

ICASSP 2017accepted0 citations

Infomax-ICA using Hessian-free optimization

Philippe Tillet, H. T. Kung, David D. Cox

Abstract

We present HF-ICA, a second-order “Hessian-free” algorithm for Infomax-ICA. Our approach achieves asymptotically quadratic convergence while retaining the memory footprint of first-order methods. Without any hyperparameter tuning, we show better convergence properties than both other approximate Newton-type methods and finely-tuned stochastic Natural Gradient Descent on EEG and fMRI data. A portable, multi-threaded and vectorized C++ implementation is made publicly available along with MATLAB and Python interfaces.

BibTeX

@inproceedings{icassp2017_infomaxicausingh,
  title = {Infomax-ICA using Hessian-free optimization},
  author = {Philippe Tillet and H. T. Kung and David D. Cox},
  booktitle = {ICASSP 2017},
  year = {2017}
}