Energy Landscapes Enable Reliable Abstention in Retrieval-Augmented Large Language Models for Healthcare
This addresses safety-critical issues in healthcare, particularly women's health, by improving abstention reliability, though it is incremental as it builds on existing energy-based and RAG methods.
The paper tackles the problem of reliable abstention in retrieval-augmented generation systems for healthcare by proposing an energy-based model that learns from 2.6M guideline-derived questions, achieving an AUROC of 0.961 on hard cases compared to 0.950 for a softmax baseline and reducing FPR@95 from 0.331 to 0.235.
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.