Hi,
Congrats on your paper and thank you for kindly sharing your research results with all of us.
May I ask if there is a way to know about the list of all models used in this research without resorting to downloading the 47 Gigabytes of data you kindly provided?
Thanks a lot in advance
I would be very appreciate if you can answer the following questions so that I can better understand your work.
Can you release the code for Figure 1 in your paper?
Should the first term in equation (3) have sigma_i? And what is the difference between f_i' and f_i?
If I understand correctly from your code, you use conv_depth_norm as overparameterization in Figure 2. I don't understand why a higher conv_depth_norm would represent overparameterization. Since if you have a shallow network, the last layers still have large conv_depth_norm. I may miss something here.
How do you choose the sparsity threshold in Eq. (6)? Should it be changed across layers and models?
For the bi-variate plot in Figure 4, what are the components for each plot? It is unclear to me how you come up with your comment from the shape of each plot, especially the second and third ones. Can you give me a clearer explanation?