I don’t understand the dilemma. Bayes theorem is a theorem that is trivially proved. There’s absolutely no question as to whether it’s correct or not. Is the question about the accuracy of the chosen prior? If you have enough samples the prior doesn’t matter because you’ll converge on the truth either way. If you don’t have enough data for a frequentist convergence then isn’t the prior the best you can do?
Thank you so much for uploading. This is gold. The epic 2009 Machine Learning Summer School also had the late Sam Roweis talks on Probabilistic Graphical Models. Wondering if you have access to them as well, and could upload. Never met Sam Roweis but having watched a few of his videos in the past, you end up feeling you actually are friends with this guy since high-school. Thank you again. J
Lex sent me here. Thanks for posting! The main Google hit for this video directs to a website so old that it expects the user to be on Windows 95. Video refuses to play on my machine.