Partial Transportability for Domain Generalization

2025-03-30

Neural Information Processing Systems

AI Summary

In domain generalization, providing provable performance guarantees for predictions on unseen target distributions remains fundamentally challenging due to the absence of target-domain data. Method: This paper proposes a tight bound estimation framework grounded in partial identifiability and transferability theory. It introduces the first general-purpose transferability estimator by adapting neural causal models (NCMs) to satisfy cross-population structural constraints and designs a gradient-based optimization algorithm for scalable inference. Contribution/Results: We establish theoretical guarantees on estimator consistency and expressive completeness. Empirical evaluation across multi-source domain settings demonstrates that our approach significantly improves the tightness and robustness of upper bounds on target-domain generalization error—particularly under black-box conditions—thereby establishing a novel paradigm for trustworthy transfer learning.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Reasoning under Uncertainty: Causality
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

A fundamental task in AI is providing performance guarantees for predictions made in unseen domains. In practice, there can be substantial uncertainty about the distribution of new data, and corresponding variability in the performance of existing predictors. Building on the theory of partial identification and transportability, this paper introduces new results for bounding the value of a functional of the target distribution, such as the generalization error of a classifier, given data from source domains and assumptions about the data generating mechanisms, encoded in causal diagrams. Our contribution is to provide the first general estimation technique for transportability problems, adapting existing parameterization schemes such Neural Causal Models to encode the structural constraints necessary for cross-population inference. We demonstrate the expressiveness and consistency of this procedure and further propose a gradient-based optimization scheme for making scalable inferences in practice. Our results are corroborated with experiments.

Problem

• Providing performance guarantees for predictions in unseen domains
• Bounding generalization error using source data and causal assumptions
• Developing scalable inference methods for transportability problems

Innovation

• Uses causal diagrams for domain generalization
• Adapts Neural Causal Models for transportability
• Proposes gradient-based optimization for scalable inference