The Price of Robustness: Stable Classifiers Need Overparameterization
This addresses the problem of understanding robustness and generalization in machine learning for researchers, providing theoretical insights into the necessity of overparameterization for stable classifiers, though it is incremental by extending prior results to discontinuous functions.
The paper tackles the relationship between overparameterization and robustness in discontinuous classifiers, showing that interpolating models with parameters roughly equal to data points must be unstable, implying substantial overparameterization is needed for high stability, with experiments indicating stability increases with model size and correlates with test performance.
The relationship between overparameterization, stability, and generalization remains incompletely understood in the setting of discontinuous classifiers. We address this gap by establishing a generalization bound for finite function classes that improves inversely with class stability, defined as the expected distance to the decision boundary in the input domain (margin). Interpreting class stability as a quantifiable notion of robustness, we derive as a corollary a law of robustness for classification that extends the results of Bubeck and Sellke beyond smoothness assumptions to discontinuous functions. In particular, any interpolating model with $p \approx n$ parameters on $n$ data points must be unstable, implying that substantial overparameterization is necessary to achieve high stability. We obtain analogous results for parameterized infinite function classes by analyzing a stronger robustness measure derived from the margin in the codomain, which we refer to as the normalized co-stability. Experiments support our theory: stability increases with model size and correlates with test performance, while traditional norm-based measures remain largely uninformative.