If correct, this looks like an important theoretical advance in understanding why and under what conditions neural nets can generalize outside their training distribution.
If correct, this looks like an important theoretical advance in understanding why and under what conditions neural nets can generalize outside their training distribution.