Order-Aware Test-Time Adaptation: Leveraging Temporal Dynamics for Robust Streaming Inference

2026-01-28

AI Summary

This work addresses the limitation of existing test-time adaptation methods, which often ignore temporal dynamics in streaming test data and consequently struggle to enhance model robustness effectively. The study formulates test-time adaptation for the first time as a gradient-free recursive Bayesian estimation problem, explicitly leveraging temporal ordering as an orthogonal supervisory signal. It introduces a lightweight, model-agnostic framework—Sequential-Aware Adaptation—that learns a dynamic transition matrix to construct a temporal prior and incorporates a likelihood-ratio gating mechanism to ensure prediction reliability under weakly structured data streams. Evaluated across diverse tasks including image classification, wearable physiological signal analysis, and language-based sentiment analysis, the proposed framework consistently outperforms baseline methods, achieving accuracy improvements of up to 6.35%.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Test-Time Adaptation (TTA) enables pre-trained models to adjust to distribution shift by learning from unlabeled test-time streams. However, existing methods typically treat these streams as independent samples, overlooking the supervisory signal inherent in temporal dynamics. To address this, we introduce Order-Aware Test-Time Adaptation (OATTA). We formulate test-time adaptation as a gradient-free recursive Bayesian estimation task, using a learned dynamic transition matrix as a temporal prior to refine the base model's predictions. To ensure safety in weakly structured streams, we introduce a likelihood-ratio gate (LLR) that reverts to the base predictor when temporal evidence is absent. OATTA is a lightweight, model-agnostic module that incurs negligible computational overhead. Extensive experiments across image classification, wearable and physiological signal analysis, and language sentiment analysis demonstrate its universality; OATTA consistently boosts established baselines, improving accuracy by up to 6.35%. Our findings establish that modeling temporal dynamics provides a critical, orthogonal signal beyond standard order-agnostic TTA approaches.

Problem

• Test-Time Adaptation
• Temporal Dynamics
• Distribution Shift
• Streaming Inference
• Order-Aware

Innovation

• Test-Time Adaptation
• Temporal Dynamics
• Recursive Bayesian Estimation
• Likelihood-Ratio Gate
• Order-Aware