Learning Conformal Abstention Policies for Adaptive Risk Management in Large Language and Vision-Language Models

TL;DR

This paper introduces a reinforcement learning-based method to adaptively set conformal prediction thresholds, improving uncertainty quantification and decision reliability in large language and vision-language models for safety-critical tasks.

Contribution

It presents a novel learnable conformal abstention approach that dynamically optimizes thresholds using reinforcement learning, surpassing static methods in accuracy and reliability.

Findings

Improves accuracy by up to 3.2% over baseline methods.

Increases AUROC for hallucination detection by 22.19%.

Reduces calibration error by 70-85%.

Abstract

Large Language and Vision-Language Models (LLMs/VLMs) are increasingly used in safety-critical applications, yet their opaque decision-making complicates risk assessment and reliability. Uncertainty quantification (UQ) helps assess prediction confidence and enables abstention when uncertainty is high. Conformal prediction (CP), a leading UQ method, provides statistical guarantees but relies on static thresholds, which fail to adapt to task complexity and evolving data distributions, leading to suboptimal trade-offs in accuracy, coverage, and informativeness. To address this, we propose learnable conformal abstention, integrating reinforcement learning (RL) with CP to optimize abstention thresholds dynamically. By treating CP thresholds as adaptive actions, our approach balances multiple objectives, minimizing prediction set size while maintaining reliable coverage. Extensive evaluations across diverse LLM/VLM benchmarks show our method outperforms Least Ambiguous Classifiers (LAC) and Adaptive Prediction Sets (APS), improving accuracy by up to 3.2%, boosting AUROC for hallucination detection by 22.19%, enhancing uncertainty-guided selective generation (AUARC) by 21.17%, and reducing calibration error by 70%-85%. These improvements hold across multiple models and datasets while consistently meeting the 90% coverage target, establishing our approach as a more effective and flexible solution for reliable decision-making in safety-critical applications. The code is available at: {https://github.com/sinatayebati/vlm-uncertainty}.