In a day crushed by health issues, I worked from my bed! I wrote up a new baby problem for my cryo-EM and deprojecting-galaxies projects: Can you infer the variance tensor for a 3-d Gaussian density blob if you only get to observe noisy 2-d projections, and you don't know any of the projection angles or offsets? Obviously the answer is yes, but the real question is how close to fully principled inference can you get, tractably, and for realistic data sets? I wrote the problem statement and half the solution (in text form); if I am confined to my bed for the rest of the week, maybe I will get to write the code too.
I also had a conversation with Marina Spivak (SCDA) about likelihoods, posteriors, marginalization, and optimization for cryo-EM, and a conversation with Ness about chemical tagging with The Cannon.