ASAP: Unsupervised Post-training with Label Distribution Shift Adaptive Learning Rate

2025-08-18

AI Summary

To address dynamic label distribution shifts and the unavailability of ground-truth labels in online deployment, this paper proposes a lightweight, unsupervised post-training adaptation method. The approach leverages only the model’s current and historical softmax outputs (i.e., soft labels), enforcing temporal consistency via contrastive modeling over output sequences and incorporating a cosine-distance-driven dynamic learning rate for single-forward-parameter updates. It requires no labels, model ensembles, storage of historical inputs, or auxiliary network architectures. Extensive experiments across multiple datasets and diverse drift patterns demonstrate significant improvements in both accuracy and convergence speed. The method achieves high computational efficiency, strong robustness to varying drift types, and seamless deployability—marking the first realization of fully real-time, soft-label-sequence-driven adaptive inference.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Learning to Search
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

In real-world applications, machine learning models face online label shift, where label distributions change over time. Effective adaptation requires careful learning rate selection: too low slows adaptation and too high causes instability. We propose ASAP (Adaptive Shift Aware Post-training), which dynamically adjusts the learning rate by computing the cosine distance between current and previous unlabeled outputs and mapping it within a bounded range. ASAP requires no labels, model ensembles, or past inputs, using only the previous softmax output for fast, lightweight adaptation. Experiments across multiple datasets and shift scenarios show ASAP consistently improves accuracy and efficiency, making it practical for unsupervised model adaptation.

Problem

• Adapting models to online label distribution shifts
• Dynamically adjusting learning rates without supervision
• Preventing instability while enabling fast adaptation

Innovation

• Dynamic learning rate adjustment via cosine distance
• Label-free adaptation using previous softmax outputs
• Lightweight unsupervised adaptation for distribution shifts