Learner Hedge Covers Sample Distortion To Achieve Generalization
——0——
This tells us that dependence alone is insufficient; its alignment with distortion is decisive. Less sample dependence need not mean better generalization.
Pedro A. Ortega@AdaptiveAgentsIn learning, the relevant perturbation is sample distortion: population loss minus empirical loss. Generalization succeeds when the learner’s hedge covers the distortion selected by its own sample dependence.
3:23 PM · May 19, 2026 · 126 Views
3:23 PM · May 19, 2026 · 139 Views