SeminKim

icon indicating a website link seminkim.github.io

icon company KAIST icon location Republic of Korea icon location Ph.D student at Vision and Learning Laboratory KAIST(School of Computing)

Results 2 issues of SeminKim

Confused with nll in vlb loss implementation.

Hi, thanks for sharing the code for your nice work! Recently I found that the training loss implementation seems a bit strange. In current implementation of L_vlb (which is for...

How does DEQ use less memory compared to explicit network?

Hi, I have a question about memory footprint of DEQ. As far as I understand, DEQ does not need to store intermediate activations, and thus able to approximate infinite-layer model...