OT Score: An OT based Confidence Score for Unsupervised Domain Adaptation

2025-05-16

AI Summary

In unsupervised domain adaptation (UDA), the absence of target-domain labels impedes reliable confidence estimation, and existing methods suffer from dual limitations in theoretical interpretability and computational feasibility. This paper proposes a confidence scoring framework grounded in semi-discrete optimal transport (OT). It is the first to theoretically incorporate the OT-induced flexible decision boundary into confidence modeling, enabling source-driven, retraining-free, dynamic confidence assessment. The framework unifies distribution alignment, pseudo-label uncertainty, and classification performance within a single principled formulation. It thus achieves rigorous theoretical foundations, high computational efficiency, and strong interpretability. On standard UDA benchmarks, the proposed confidence scores significantly outperform existing metrics; when integrated into confidence-based filtering, they consistently improve classification accuracy.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Large-scale security measurements

Abstract

We address the computational and theoretical limitations of existing distributional alignment methods for unsupervised domain adaptation (UDA), particularly regarding the estimation of classification performance and confidence without target labels. Current theoretical frameworks for these methods often yield computationally intractable quantities and fail to adequately reflect the properties of the alignment algorithms employed. To overcome these challenges, we introduce the Optimal Transport (OT) score, a confidence metric derived from a novel theoretical analysis that exploits the flexibility of decision boundaries induced by Semi-Discrete Optimal Transport alignment. The proposed OT score is intuitively interpretable, theoretically rigorous, and computationally efficient. It provides principled uncertainty estimates for any given set of target pseudo-labels without requiring model retraining, and can flexibly adapt to varying degrees of available source information. Experimental results on standard UDA benchmarks demonstrate that classification accuracy consistently improves by identifying and removing low-confidence predictions, and that OT score significantly outperforms existing confidence metrics across diverse adaptation scenarios.

Problem

• Addresses limitations in unsupervised domain adaptation confidence estimation
• Introduces OT score for interpretable, efficient performance assessment
• Improves classification accuracy by filtering low-confidence predictions

Innovation

• Optimal Transport score for confidence estimation
• Semi-Discrete Optimal Transport alignment flexibility
• No model retraining for uncertainty estimates