On the Learnability of Test-Time Adaptation: A Recovery Complexity Perspective

2026-05-27

AI Summary

This work addresses the lack of a learnability theory in test-time adaptation (TTA) that simultaneously accounts for adaptation objectives, continuous distribution shifts, and information constraints. We propose the first theoretical framework for TTA learnability, introducing the notions of $(\varepsilon,\delta)$-recovery complexity and $(\varepsilon,\rho)$-TTA learnability, and unify the characterization of both gradual and abrupt distribution shifts through a discrete proxy model. By integrating recovery complexity measures, discretization of non-stationary streams, probabilistic risk control, and information-theoretic analysis, we derive matching upper and lower bounds on recovery complexity. This reveals an intrinsic trade-off between adaptation capability and available information, providing long-term reliability guarantees for TTA and filling a critical gap left by existing regret-based theoretical approaches.

Abstract

Test-time adaptation (TTA) aims to adapt models to maintain reliable performance on non-stationary test streams without requiring labeled data. Despite its empirical success, the learnability of TTA under non-stationary streams remains unexplored. A key challenge is the lack of a principled theoretical framework that simultaneously aligns with the TTA objective and captures both continuously evolving distribution shifts and intrinsic information constraints. To address this gap, we propose the first theoretical framework for studying the learnability of TTA and introduce $(ε,δ)$-Recovery Complexity and $(ε,ρ)$-TTA Learnability. Recovery complexity measures the post-shift time needed to maintain excess risk below a target level with high probability, and is further extended to TTA learnability, which measures the long-term reliability of TTA. Within this framework, we introduce a novel discrete surrogate for non-stationary test streams, enabling a unified and tractable analysis of both gradual and abrupt shifts. We derive order-wise matching lower and upper bounds on recovery complexity, revealing fundamental limits of TTA and an intrinsic adaptivity-information trade-off. These results provide unified learnability guarantees for TTA that complement regret-based analyses.

Problem

• test-time adaptation
• non-stationary streams
• distribution shift
• learnability
• recovery complexity

Innovation

• test-time adaptation
• recovery complexity
• distribution shift
• learnability
• non-stationary streams