PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation

2024-03-15

arXiv.org

AI Summary

Existing continual test-time adaptation (CTTA) methods rely on error-prone pseudo-labels and yield inaccurate domain shift estimation, leading to performance degradation in dynamic environments. To address this, we propose a pseudo-label-free CTTA framework. First, we introduce a novel uncertainty-driven dynamic layer selection mechanism that quantifies prediction uncertainty under the uniform distribution via the gradient magnitude of KL divergence—enabling adaptive layer freezing without pseudo-labels. Second, we design a layer-wise domain shift approximation based on parameter sensitivity, enabling hierarchical learning rate adaptation. Our method eliminates pseudo-label dependency, significantly improving adaptation robustness and stability. Extensive experiments on CIFAR-10C, CIFAR-100C, and ImageNet-C demonstrate state-of-the-art performance: average error rates are reduced by up to 12.7% compared to prior art, confirming substantial gains in generalization and continual adaptability.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Real-world vision models in dynamic environments face rapid shifts in domain distributions, leading to decreased recognition performance. Using unlabeled test data, continuous test-time adaptation (CTTA) directly adjusts a pre-trained source discriminative model to these changing domains. A highly effective CTTA method involves applying layer-wise adaptive learning rates for selectively adapting pre-trained layers. However, it suffers from the poor estimation of domain shift and the inaccuracies arising from the pseudo-labels. This work aims to overcome these limitations by identifying layers for adaptation via quantifying model prediction uncertainty without relying on pseudo-labels. We utilize the magnitude of gradients as a metric, calculated by backpropagating the KL divergence between the softmax output and a uniform distribution, to select layers for further adaptation. Subsequently, for the parameters exclusively belonging to these selected layers, with the remaining ones frozen, we evaluate their sensitivity to approximate the domain shift and adjust their learning rates accordingly. We conduct extensive image classification experiments on CIFAR-10C, CIFAR-100C, and ImageNet-C, demonstrating the superior efficacy of our method compared to prior approaches.

Problem

• Continuous Time Transfer Adaptation
• Environmental Changes
• Pseudo-label Accuracy

Innovation

• Palm
• Uncertainty Estimation
• Adaptive Learning Rate