generalization-theory-in-fundamental-revision
IN derived (depth 2)
Created 2026-06-21T10:09:45+00:00 · Reviewed 2026-06-21T15:37:01+00:00
ML generalization theory is undergoing fundamental revision on two independent fronts — the No Free Lunch theorem established that no universal best model exists (killing the quest for a single optimal algorithm), while double descent and benign overfitting overturned the classical U-shaped bias-variance curve (killing the traditional model selection heuristic) — leaving the field without a reliable theoretical guide to practice.
Justifications
SL — Both model universality and the bias-variance U-curve are refuted, removing the two pillars of classical model selection
Antecedents (all must be IN):
- IN classical-generalization-theory-overturned — Classical generalization theory — the U-shaped bias-variance tradeoff — has been overturned by two empirical phenomena: double descent shows test error decreasing again far past the interpolation threshold, and benign overfitting shows perfect training fit coexisting with good generalization in overparameterized regimes.
- IN no-universal-optimal-model — No universally optimal ML model exists: the No Free Lunch theorem establishes this impossibility in principle, and the bias-variance decomposition reveals the mechanism — any fixed model trades bias against variance depending on the problem.
Dependents
These beliefs depend on this one:
- IN generalization-framework-unified-yet-under-revision — The bias-variance framework simultaneously unifies all classical generalization techniques (regularization, ensembles, dropout) AND is being fundamentally revised by double descent and benign overfitting — the organizing principle works as engineering guidance but its theoretical foundations are shifting beneath it.
- IN theoretical-crisis-parallels-practical-fragility — ML's theoretical and practical reliability crises are parallel and reinforcing — generalization theory is in fundamental revision as double descent and benign overfitting undermine the classical framework, while deployed neural networks face two independent failure classes (adversarial vulnerability, algorithmic bias) that the revising theory cannot yet predict or prevent.