The index

Explore the frontier

2 entries matching the current filters. Sorted by recently added.

Loss follows a power law in compute, data and parameters. No theory predicts the exponent from the data distribution.

For linear models the answer is a clean minimum-norm statement. For anything deeper, there is no general theory.