Oracle inequalities for computationally adaptive model selection
This work addresses the practical challenge of computationally efficient model selection for machine learning practitioners, offering a novel integration of computational aspects into theoretical analysis.
The paper tackles the problem of model selection under computational constraints by introducing a framework and algorithms that trade off estimation error, approximation error, and computational effort. It provides oracle inequalities showing that the risk of the selected model is close to that of the optimal function class, with specific bounds on performance degradation.
We analyze general model selection procedures using penalized empirical loss minimization under computational constraints. While classical model selection approaches do not consider computational aspects of performing model selection, we argue that any practical model selection procedure must not only trade off estimation and approximation error, but also the computational effort required to compute empirical minimizers for different function classes. We provide a framework for analyzing such problems, and we give algorithms for model selection under a computational budget. These algorithms satisfy oracle inequalities that show that the risk of the selected model is not much worse than if we had devoted all of our omputational budget to the optimal function class.