@@samprasdsouza6993 It is just similar to how you would want to handle over fitting in general, being skeptical ;) So the assumption that Bayesian doesn't over-fit is not valid. It is actually just so crystal clear and meaningful with Bayesian construct why, what and the how! uniform distribution is the most max entropy but it sees all the data equally likely no skepticism a very narrow one is a very skeptical prior. you need to solve the trade off! so the regularization is already in the prior selection stage! Ridge in ML wording is a Gaussian prior, an lasso is Laplacian one
Pretty cool lecture! Thank you.
Great talk!
Bayesian do over-fit a maximum entropy prior will tend to over-fit and encode all the sample noise:)
yes, I recently faced a similar problem. any way we can overcome this ?
@@samprasdsouza6993 It is just similar to how you would want to handle over fitting in general, being skeptical ;) So the assumption that Bayesian doesn't over-fit is not valid. It is actually just so crystal clear and meaningful with Bayesian construct why, what and the how! uniform distribution is the most max entropy but it sees all the data equally likely no skepticism a very narrow one is a very skeptical prior. you need to solve the trade off! so the regularization is already in the prior selection stage! Ridge in ML wording is a Gaussian prior, an lasso is Laplacian one
@@mohamedibrahim1836 sorry but this solution went over my head. I am beginner in this area can you recommend me relevant papers/articles
@@mohamedibrahim1836 sorry this solution went over my head. can you recommend me relevant research papers/articles
@@mohamedibrahim1836 sorry but I was not able to fully comprehend your response. can you please suggest me relevant papers/articles
Where did you got the idea of pruning based on the probability of being zero? Do you have some references¿
Thanks!
I also have been looking for similar references recently;
one paper to check out; the BBB paper: proceedings.mlr.press/v37/blundell15.html
I really should have put all my money into Bitcoin back then.