Energy Landscapes Enable Reliable Abstention in Retrieval-Augmented Large Language Models for Healthcare

2025-08-31

AI Summary

Medical retrieval-augmented generation (RAG) systems pose safety risks in high-stakes domains—e.g., women’s health—when generating erroneous responses. Method: This paper proposes an energy-based abstention mechanism that models intrinsic uncertainty of query-answer pairs within a dense semantic space. By constructing a smooth energy landscape, the method integrates calibrated softmax contrastive learning, k-nearest-neighbor density estimation, and controlled negative sampling—departing from conventional probabilistic confidence scores. Contribution/Results: The approach significantly improves abstention discrimination for in-distribution approximate queries. On hard abstention evaluation, it achieves an AUROC of 0.961 and a false positive rate at 95% true positive rate (FPR@95) of 0.235—substantially outperforming baselines. These results demonstrate the efficacy and robustness of energy scoring for safety-critical medical AI applications.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Reliable abstention is critical for retrieval-augmented generation (RAG) systems, particularly in safety-critical domains such as women's health, where incorrect answers can lead to harm. We present an energy-based model (EBM) that learns a smooth energy landscape over a dense semantic corpus of 2.6M guideline-derived questions, enabling the system to decide when to generate or abstain. We benchmark the EBM against a calibrated softmax baseline and a k-nearest neighbour (kNN) density heuristic across both easy and hard abstention splits, where hard cases are semantically challenging near-distribution queries. The EBM achieves superior abstention performance abstention on semantically hard cases, reaching AUROC 0.961 versus 0.950 for softmax, while also reducing FPR@95 (0.235 vs 0.331). On easy negatives, performance is comparable across methods, but the EBM's advantage becomes most pronounced in safety-critical hard distributions. A comprehensive ablation with controlled negative sampling and fair data exposure shows that robustness stems primarily from the energy scoring head, while the inclusion or exclusion of specific negative types (hard, easy, mixed) sharpens decision boundaries but is not essential for generalisation to hard cases. These results demonstrate that energy-based abstention scoring offers a more reliable confidence signal than probability-based softmax confidence, providing a scalable and interpretable foundation for safe RAG systems.

Problem

• Ensuring reliable abstention in healthcare RAG systems
• Handling semantically challenging near-distribution medical queries
• Providing safer confidence scoring than probability-based methods

Innovation

• Energy-based model learns smooth energy landscape
• EBM achieves superior abstention on hard cases
• Energy scoring provides reliable confidence signal