BD-Merging: Bias-Aware Dynamic Model Merging with Evidence-Guided Contrastive Learning

TL;DR

BD-Merging introduces a bias-aware, uncertainty modeling framework for model merging in multi-task learning, improving robustness and reliability under distribution shifts through evidence-guided contrastive learning.

Contribution

It proposes a novel bias-aware unsupervised model merging method that explicitly models uncertainty and uses evidence-guided contrastive learning to enhance robustness under distribution shifts.

Findings

Outperforms state-of-the-art MM methods in diverse tasks.

Effectively mitigates bias and improves generalization under distribution shift.

Achieves superior robustness and effectiveness in empirical evaluations.

Abstract

Model Merging (MM) has emerged as a scalable paradigm for multi-task learning (MTL), enabling multiple task-specific models to be integrated without revisiting the original training data. Despite recent progress, the reliability of MM under test-time distribution shift remains insufficiently understood. Most existing MM methods typically assume that test data are clean and distributionally aligned with both the training and auxiliary sources. However, this assumption rarely holds in practice, often resulting in biased predictions with degraded generalization. To address this issue, we present BD-Merging, a bias-aware unsupervised model merging framework that explicitly models uncertainty to achieve adaptive reliability under distribution shift. First, BD-Merging introduces a joint evidential head that learns uncertainty over a unified label space, capturing cross-task semantic dependencies in MM. Second, building upon this evidential foundation, we propose an Adjacency Discrepancy Score (ADS) that quantifies evidential alignment among neighboring samples. Third, guided by ADS, a discrepancy-aware contrastive learning mechanism refines the merged representation by aligning consistent samples and separating conflicting ones. Combined with general unsupervised learning, this process trains a debiased router that adaptively allocates task-specific or layer-specific weights on a per-sample basis, effectively mitigating the adverse effects of distribution shift. Extensive experiments across diverse tasks demonstrate that BD-Merging achieves superior effectiveness and robustness compared to state-of-the-art MM baselines.