search

chl8856 / DeepIMV

A Variational Information Bottleneck Approach to Multi-Omics Data Integration
21 stars 12 forks source link

Does the LOSS_CONSISTENCY help? #2

Open John-zf opened 3 years ago

John-zf commented 3 years ago

I found the LOSS_CONSISTENCY in your code, however it is not appeared in your paper. I have run the code on my own dataset, but it seems that the LOSS_CONSISTENCY is not stable. Does the LOSS_CONSISTENCY mean to minimize the difference the joint distribution and the marginal distributions of the multiple variational encoders?

chl8856 commented 3 years ago

Hello, LOSS_CONSISTENCY was first intended to improve the stability of the learning by encouraging the network to share information across different views. (This is achieved by constraining the joint and the marginal distributions similar). However, it was not helpful in the current form, and thus, not used for experiments in the paper.

John-zf commented 3 years ago

Hello, LOSS_CONSISTENCY was first intended to improve the stability of the learning by encouraging the network to share information across different views. (This is achieved by constraining the joint and the marginal distributions similar). However, it was not helpful in the current form, and thus, not used for experiments in the paper.

thx!