PDRL: Post-hoc Descriptor-based Residual Learning for Uncertainty-Aware Machine Learning Potentials

2025-09-02

AI Summary

Existing machine-learned interatomic potentials (MLIPs) lack efficient uncertainty quantification (UQ) capabilities without costly retraining. To address this, we propose PDRL—a descriptor-based, post-hoc residual learning framework that requires no retraining of the base MLIP. PDRL leverages atomistic environment descriptors extracted by graph neural networks to directly model the distribution of prediction residuals, enabling rapid, low-overhead UQ for any pre-trained MLIP. Its core innovation lies in decoupling UQ from model training as a standalone post-processing task, thereby preserving the original model’s accuracy. Extensive experiments across multiple benchmark datasets demonstrate that PDRL significantly outperforms state-of-the-art UQ methods in error calibration and uncertainty estimation fidelity, while incurring negligible additional computational cost. PDRL exhibits strong generalizability and plug-and-play compatibility with diverse MLIP architectures.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Knowledge Representation and Reasoning: Description Logics

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: ML for personalized search and recommendations

Abstract

Ensemble method is considered the gold standard for uncertainty quantification (UQ) for machine learning interatomic potentials (MLIPs). However, their high computational cost can limit its practicality. Alternative techniques, such as Monte Carlo dropout and deep kernel learning, have been proposed to improve computational efficiency; however, some of these methods cannot be applied to already trained models and may affect the prediction accuracy. In this paper, we propose a simple and efficient post-hoc framework for UQ that leverages the descriptor of a trained graph neural network potential to estimate residual errors. We refer to this method as post-hoc descriptor-based residual-based learning (PDRL). PDRL models the discrepancy between MLIP predictions and ground truth values, allowing these residuals to act as proxies for prediction uncertainty. We explore multiple variants of PDRL and benchmark them against established UQ methods, evaluating both their effectiveness and limitations.

Problem

• Reducing computational cost of uncertainty quantification in ML potentials
• Enabling post-hoc uncertainty estimation without retraining models
• Improving accuracy-efficiency tradeoff in interatomic potential uncertainty methods

Innovation

• Post-hoc descriptor-based residual learning framework
• Leverages graph neural network descriptor
• Models discrepancy as uncertainty proxies