Briefly skimmed the paper and I'm not a ML or math dude but familiar with the double descent problem. I wonder if there's a good way to measure information con

Briefly skimmed the paper and I'm not a ML or math dude but familiar with the double descent problem. I wonder if there's a good way to measure information content in models and how it scales with model parameters, if there are any invariants, scaling law's that arise etc. Reminds me of renormalization group methods which could be applied to a space of models, etc... Again, I'm no expert in any of this stuff, just arm-chairing away.