helsinki1.jpg

In this project, the goal was to incorporate explicit control of representation sparseness into the framework of NMF. In a sense, we are combining the ideas of ICA and NMF.

An important aspect to consider in most ICA-based analysis is the stability of the results with respect to small changes in the dataset. One wants to know if the components are reliable in the sense that different random subsets of the data produce roughly the same results. We have developed a method, called ICASSO, for answering exactly this question.

For more information, as well as the full code package, see the ICASSO home page.

Independent component analysis is a fundamental data analysis tool, developed in the 1990's. One of the most influential contributions of our group is the FastICA algorithm for performing ICA. More information, as well as a complete Matlab code package, is available at the FastICA home page.

Patrik Hoyer, Aapo Hyvärinen, and Michael Gutmann

Previous studies (see our other projects) have shown how many basic properties of the primary visual cortex, such as the receptive fields of simple and complex cells and the spatial organization (topography) of the cells, can be understood as efficient coding of natural images. In this project we extend the framework by considering how the responses of complex cells (when fed with natural image input) could be sparsely represented by a higher-order neural layer.

The fundamental results relating natural stimulus statistics to the structure and functionality of neurons in the primary visual cortex were obtained with static image data. These results suggested that the defining property of the static neural code at the simple-cell level is sparseness / statistical independence.