Characterization of Excess Risk for Locally Strongly Convex Population Risk

The core problem in machine learning is obtaining a model that generalizes well on unseen test data. The excess risk decides the model's performance on these unseen data, and it can be decomposed