Beyond Prompt: Fine-grained Simulation of Cognitively Impaired Standardized Patients via Stochastic Steering

arXiv cs.AI / 4/15/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes StsPatient, a method to simulate standardized patients with cognitive impairments for scalable and ethical clinical training.
It moves beyond discrete prompt engineering by learning fine-grained, domain-specific steering vectors from instruction–response contrastive pairs.
The approach adds Stochastic Token Modulation (STM) to control the intervention probability, aiming to improve stability versus conventional vector-based interventions.
Experiments report improved clinical authenticity and better controllability of impairment severity compared with baseline methods.

Abstract

Simulating Standardized Patients with cognitive impairment offers a scalable and ethical solution for clinical training. However, existing methods rely on discrete prompt engineering and fail to capture the heterogeneity of deficits across varying domains and severity levels. To address this limitation, we propose StsPatient for the fine-grained simulation of cognitively impaired patients. We innovatively capture domain-specific features by extracting steering vectors from contrastive pairs of instructions and responses. Furthermore, we introduce a Stochastic Token Modulation (STM) mechanism to regulate the intervention probability. STM enables precise control over impairment severity while mitigating the instability of conventional vector methods. Comprehensive experiments demonstrate that StsPatient significantly outperforms baselines in both clinical authenticity and severity controllability.