Have to head out — thank you!
Correct me if I am wrong, incase of overparametrized NNs , wouldn't the loss have a double descent?
Could you remind us which papers each of the sections of your talk was based on? Thanks.
I think this is the first one https://arxiv.org/pdf/2006.15812v1.pdf
actually I pasted the wrong link, sorry xD
Hardness for unit norm input and unit norm weightd