The day started with a journal-club talk by Eftychios Pnevmatikakis (SCDA) about spike sorting and neuron deblending in calcium imaging. This is a kind of video-rate imaging in which you see the activity in the neurons of a live brain by fluorescence. Every technique he uses in these contexts is a technique used in astronomical image analysis, from sparse and non-negative matrix factorizations to MCMC with reversible jump. This was yet another reminder that astronomy and microscopy are very similar fields, technically.
Andy Casey and I did get our factor of 150 speed-up by including the analytic derivatives! We still have concerns about convergence, but you can do a lot more experiments along those lines when your code is 150 times faster. We switched gears today to continuum normalization. For this purpose, we built a simple linear least-square fitter with light ridge regularization (L2) to keep the matrix well conditioned. We started downloading all the individual-exposure files, because the whole point of our regularized version of The Cannon is that it will work well at much lower signal-to-noise than traditional methods.