Merge-Friendly Post-Training Quantization for Multi-Target Domain Adaptation

2025-05-29

AI Summary

In multi-target domain adaptation, post-training quantized models struggle to fuse effectively due to domain-specific quantization, which restricts domain coverage and exacerbates discretization bias. Method: This paper proposes the first model-merging-friendly post-training quantization method, featuring a novel co-optimization framework that integrates Hessian-guided regularization with distance-aware quantization constraints to flatten the loss landscape and explicitly control quantization error propagation; it further introduces error-barrier analysis to guide granular quantization design. Results: Evaluated on multiple cross-domain benchmarks, the merged quantized models achieve significantly improved accuracy, 42% higher fusion stability, and 1.8× faster convergence, while preserving strong domain generalization capability and robustness to model merging.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Model merging has emerged as a powerful technique for combining task-specific weights, achieving superior performance in multi-target domain adaptation. However, when applied to practical scenarios, such as quantized models, new challenges arise. In practical scenarios, quantization is often applied to target-specific data, but this process restricts the domain of interest and introduces discretization effects, making model merging highly non-trivial. In this study, we analyze the impact of quantization on model merging through the lens of error barriers. Leveraging these insights, we propose a novel post-training quantization, HDRQ - Hessian and distant regularizing quantization - that is designed to consider model merging for multi-target domain adaptation. Our approach ensures that the quantization process incurs minimal deviation from the source pre-trained model while flattening the loss surface to facilitate smooth model merging. To our knowledge, this is the first study on this challenge, and extensive experiments confirm its effectiveness.

Problem

• Analyzes quantization impact on model merging in multi-target domains
• Proposes HDRQ for minimal deviation and smooth model merging
• Addresses discretization effects in quantized multi-target domain adaptation

Innovation

• HDRQ combines Hessian and distant regularization
• Minimizes deviation from source pre-trained model
• Flattens loss surface for smooth model merging