In my work on the learning2learn collaboration, we are interested in examining changes in multi-channel recordings across episodes of learning and the larger establishing of learning schemata. In a (possibly criminal) simplication of the idea (and based on revisiting this Cunningham and Yu review): one of the ways to examine this process is by looking at the ‘intrinsic’ dimensionality of the data and the trajectory of neural activity through the lower-dimensional representation with behavioral changes. John Ferre, who is a graduate student here at UW, has started computing these things for a relatively small amount of data, on our PanNeuro deployment. So far, we’ve been setting up a pipeline for spectral decomposition and for construction of a large covariance matrix, as the first step towards a distributed PCA of multi-channel data.