search

Linear95 / CLUB

Code for ICML2020 paper - CLUB: A Contrastive Log-ratio Upper Bound of Mutual Information
311 stars 39 forks source link

Questions about the bound #24

Open DZN-Research opened 1 year ago

DZN-Research commented 1 year ago

Dear author, Thanks for your great work and implementation. I have a similar network architecture like the MI Minimization in Domain Adaptation in the paper. When I use the CLUB to minimize the mutual information between content features and domain features, the bound still reduces at the first few epochs. However, it can reduces to less than 0 at an epoch and becomes larger and larger at the following epochs. Have you encounterd such situation? Do you have any advice for this? Thank you very much!

trailblazer-bit commented 11 months ago

Hi, I also encountered a similar problem. How did you solve it?

liutaocode commented 10 months ago

same issue

Linear95 commented 10 months ago

Hi there, sorry for not noticing your questions in a closed issue. Please check some of my suggestions in #17