Deep linear networks for regression are implicitly regularized towards flat minima

Connections are not prepared for this record yet, or its source metadata has changed. The original record remains available while background snapshots are rebuilt.