principal component analysis

Principal component analysis (PCA) aims to describe whole-brain spatio-temporal responses using a low-dimensional factorization, with spatial and temporal factors that best capture the joint dynamics. The temporal factors capture common profiles of temporal response, and the corresponding spatial factors show to what extent individual voxels exhibit those temporal profiles. This relationship can be visualized in a map using color to represent the shape of the response, and brightness to represent the strength. To help clarify this relationship, the points on the left show the low-dimensional embedding; each point represents a neural response. By mousing over them you can see the time series of that response as reconstructed using the first two components. PCA is implemented using the singular value decomposition, with two choices for the algorithm. Visualization in collaboration with Matt Conlen.

See a full analysis in an iPython notebook