Rare event modeling with self-regularized normalizing flows: what can we learn from a single failure?

2025-02-28

AI Summary

This paper addresses the challenge of modeling and root-cause diagnosis for safety-critical failures in autonomous systems (e.g., autonomous driving, air traffic scheduling), where failure data are extremely scarce—often limited to mere seconds of pre-failure observations. To tackle this, we propose CalNF, the first self-regularized normalizing flow framework. CalNF integrates invertible neural networks, Bayesian posterior estimation, and self-supervised uncertainty-aware regularization; it implicitly calibrates priors to dynamically balance overfitting and underfitting, enabling reliable posterior distribution learning from single-sample failure instances. Evaluated on multi-source sparse failure modeling tasks, CalNF achieves state-of-the-art performance. Notably, it is the first method to successfully reconstruct and interpret the deep causal chain underlying the 2022 Southwest Airlines scheduling crisis via generative inversion, establishing a novel paradigm for interpretable modeling of rare failures.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Causality
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Increased deployment of autonomous systems in fields like transportation and robotics have seen a corresponding increase in safety-critical failures. These failures can be difficult to model and debug due to the relative lack of data: compared to tens of thousands of examples from normal operations, we may have only seconds of data leading up to the failure. This scarcity makes it challenging to train generative models of rare failure events, as existing methods risk either overfitting to noise in the limited failure dataset or underfitting due to an overly strong prior. We address this challenge with CalNF, or calibrated normalizing flows, a self-regularized framework for posterior learning from limited data. CalNF achieves state-of-the-art performance on data-limited failure modeling and inverse problems and enables a first-of-a-kind case study into the root causes of the 2022 Southwest Airlines scheduling crisis.

Problem

• Modeling rare safety-critical failures with limited data.
• Addressing overfitting and underfitting in generative models.
• Investigating root causes of rare events like airline crises.

Innovation

• Self-regularized normalizing flows for rare events
• Calibrated normalizing flows (CalNF) for limited data
• Improved failure modeling with state-of-the-art performance