Question about sparse resnet

[junleiz]

Hello:
Thank you for your job. I am interested in your sparse resnet. I am not quite sure how do you implement it. Or you just replace the "concat" operations with "add"?

[Spinkoo]

I do indeed share the same question, Like since the shortest path of the gradient is O(1) in a resnet due to the sum of all the previous feature maps (kind of implicitly), how exactly is the path then extended using sparsenet ?