CURVE: Learning Causality-Inspired Invariant Representations for Robust Scene Understanding via Uncertainty-Guided Regularization

2026-01-28

AI Summary

This work addresses the vulnerability of scene graph generation to spurious correlations, which undermines out-of-distribution generalization. To this end, the authors propose the CURVE framework, which disentangles environment-invariant interaction dynamics from environment-specific variations through variational uncertainty modeling and uncertainty-guided structural regularization, thereby constructing a sparse and domain-stable topological structure. Additionally, a causally inspired prototype-conditioned debiasing mechanism is introduced to suppress high-variance, environment-specific relations by leveraging predictive uncertainty. The method significantly improves generalization performance in zero-shot transfer and low-data simulation-to-real (sim-to-real) settings, while also providing reliable uncertainty estimates that enable risk prediction under distributional shifts.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Scene graphs provide structured abstractions for scene understanding, yet they often overfit to spurious correlations, severely hindering out-of-distribution generalization. To address this limitation, we propose CURVE, a causality-inspired framework that integrates variational uncertainty modeling with uncertainty-guided structural regularization to suppress high-variance, environment-specific relations. Specifically, we apply prototype-conditioned debiasing to disentangle invariant interaction dynamics from environment-dependent variations, promoting a sparse and domain-stable topology. Empirically, we evaluate CURVE in zero-shot transfer and low-data sim-to-real adaptation, verifying its ability to learn domain-stable sparse topologies and provide reliable uncertainty estimates to support risk prediction under distribution shifts.

Problem

• scene graph
• spurious correlations
• out-of-distribution generalization
• domain shift
• robust scene understanding

Innovation

• causality-inspired learning
• uncertainty-guided regularization
• invariant representation
• scene graph debiasing
• domain generalization