ElimPCL: Eliminating Noise Accumulation with Progressive Curriculum Labeling for Source-Free Domain Adaptation

2025-03-31

AI Summary

In source-free domain adaptation (SFDA), pre-trained source models generate highly uncertain pseudo-labels for hard samples in the target domain, leading to noise accumulation and erroneous propagation in feature space. To address this, we propose ElimPCL: (1) a prototype-consistency-based progressive curriculum labeling mechanism that dynamically selects high-confidence samples; and (2) a novel dual-path MixUp feature augmentation strategy that fuses reliable and hard samples in feature space, enhancing discriminability of hard samples while suppressing noise interference. ElimPCL operates without access to source data or labels, requiring only a single forward pass on the target domain. Evaluated on multiple SFDA benchmarks, it consistently outperforms state-of-the-art methods, achieving up to a 3.4% absolute accuracy gain. The framework effectively mitigates both the accumulation and diffusion of pseudo-label noise.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Source-Free Domain Adaptation (SFDA) aims to train a target model without source data, and the key is to generate pseudo-labels using a pre-trained source model. However, we observe that the source model often produces highly uncertain pseudo-labels for hard samples, particularly those heavily affected by domain shifts, leading to these noisy pseudo-labels being introduced even before adaptation and further reinforced through parameter updates. Additionally, they continuously influence neighbor samples through propagation in the feature space.To eliminate the issue of noise accumulation, we propose a novel Progressive Curriculum Labeling (ElimPCL) method, which iteratively filters trustworthy pseudo-labeled samples based on prototype consistency to exclude high-noise samples from training. Furthermore, a Dual MixUP technique is designed in the feature space to enhance the separability of hard samples, thereby mitigating the interference of noisy samples on their neighbors.Extensive experiments validate the effectiveness of ElimPCL, achieving up to a 3.4% improvement on challenging tasks compared to state-of-the-art methods.

Problem

• Reducing noise in pseudo-labels for SFDA
• Handling domain shifts in hard samples
• Preventing noise propagation in feature space

Innovation

• Progressive Curriculum Labeling filters noisy pseudo-labels
• Dual MixUP enhances hard sample separability
• Prototype consistency excludes high-noise samples