Human-in-the-Loop Multi-Agent Ventilator Decision Support with Contextual Bandit Preference Learning
A new paper presents a Human-in-the-Loop Multi-Agent Ventilator Decision Support System (VDSS) that utilizes contextual bandit preference learning. This system aims to improve ventilator decision-making by adapting to clinician preferences and enhancing collaboration between human and AI. The study indicates that VDSS can lead to higher recommendation acceptability and fewer iterations to reach a satisfactory plan in clinical settings.
- ▪The VDSS framework coordinates modular decision components through structured interfaces.
- ▪It performs online preference adaptation based on clinician-specific decisions.
- ▪Retrospective reviews show improved interaction stability and reduced planning iterations.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23320 (cs) [Submitted on 22 May 2026] Title:Human-in-the-Loop Multi-Agent Ventilator Decision Support with Contextual Bandit Preference Learning Authors:Sijia Li, Xiaoyu Tan, Qixing Wang, Weiyi Zhao, Chen Zhan, Teqi Hao, Xuemin Wang, Lei Gu, Roland Eils, Xihe Qiu View a PDF of the paper titled Human-in-the-Loop Multi-Agent Ventilator Decision Support with Contextual Bandit Preference Learning, by Sijia Li and 9 other authors View PDF HTML (experimental) Abstract:Ventilator decision support requires sequential decisions that track evolving physiology and disease trajectories while respecting safety boundaries and clinician specific tuning styles.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
