Machine learning (ML) models are highly flexible but face fundamental limits when learning from finite data. We show that a universal lower bound—the Limits-to-Learning Gap (LLG)—captures the unavoidable gap between the true (population) benchmark and linear ML models.
LQG 2026/01/15 – 18:30 – Limits to (Machine) Learning – Semyon Malamud – IP *OLO* Hybrid
read more
