Anirudh Cowlagi

Results 2 comments of Anirudh Cowlagi

As a follow up, here is the network I am using: ``` class allcnn_t(nn.Module): def __init__(self, c1=16, c2= 32): super().__init__() d = 0 def convbn(ci,co,ksz,s=1,pz=0): return nn.Sequential( nn.Conv2d(ci,co,ksz,stride=s,padding=pz), nn.ReLU(), nn.BatchNorm2d(co))...

Thanks this is helpful -- I will adopt the second approach! A followup related to eigendecompositions using the KFAC representation. I understand that the eigenvalues in this representation can be...