The benefit of multitask representation learning

Maurer, Pontil, and Romera-Paredes derived finite-sample excess-risk bounds for multi-task representation learning. We carry their shared-representation insight into dalla2026between and recast it through the separation condition for evolutionary perception.

Cited alongside Baxter in riva2026task for the sample-complexity benefit of multi-task representation learning, again as a contrast with our identification of an exact loss-induced quotient.