We have ImageNet experiments in Section 4 of the full paper: https://arxiv.org/abs/2010.08127
This is the exact comparison we make in the paper. We have subsampled ImageNet experiments as well; see Figure 3 in the full paper: https://arxiv.org/abs/2010.08127
Oh yes, absolutely. We only make claims for "natural distributions and models" (^). It's almost certainly possible to break this by pathological choice of a data distribution or model/initialization/optimization scheme.…
Hi, we define the interpolation threshold in Section 2 of the full paper (https://arxiv.org/abs/1912.02292) as the point when the "Effective Model Complexity" = # of train samples. Where the "Effective Model Complexity…
We have ImageNet experiments in Section 4 of the full paper: https://arxiv.org/abs/2010.08127
This is the exact comparison we make in the paper. We have subsampled ImageNet experiments as well; see Figure 3 in the full paper: https://arxiv.org/abs/2010.08127
Oh yes, absolutely. We only make claims for "natural distributions and models" (^). It's almost certainly possible to break this by pathological choice of a data distribution or model/initialization/optimization scheme.…
Hi, we define the interpolation threshold in Section 2 of the full paper (https://arxiv.org/abs/1912.02292) as the point when the "Effective Model Complexity" = # of train samples. Where the "Effective Model Complexity…