The Over-Certainty Phenomenon in Modern Test-Time Adaptation Algorithms
This addresses the calibration issue in test-time adaptation for neural networks, which is crucial for reliable deployment in real-world scenarios with domain shifts, though it is incremental as it builds on existing adaptation methods.
The paper tackles the problem of neural networks becoming over-confident when adapting to unfamiliar data during test-time, which can lead to misplaced trust. It proposes a certainty regularizer that dynamically adjusts pseudo-label confidence, achieving state-of-the-art performance in calibration metrics like Expected Calibration Error and Negative Log Likelihood while maintaining accuracy.
When neural networks are confronted with unfamiliar data that deviate from their training set, this signifies a domain shift. While these networks output predictions on their inputs, they typically fail to account for their level of familiarity with these novel observations. Prevailing works navigate test-time adaptation with the goal of curtailing model entropy, yet they unintentionally produce models that struggle with sub-optimal calibration-a dilemma we term the over-certainty phenomenon. This over-certainty in predictions can be particularly dangerous in the setting of domain shifts, as it may lead to misplaced trust. In this paper, we propose a solution that not only maintains accuracy but also addresses calibration by mitigating the over-certainty phenomenon. To do this, we introduce a certainty regularizer that dynamically adjusts pseudo-label confidence by accounting for both backbone entropy and logit norm. Our method achieves state-of-the-art performance in terms of Expected Calibration Error and Negative Log Likelihood, all while maintaining parity in accuracy.