Multi-Agent Norm Perception and Induction in Distributed Healthcare

TL;DR

The paper tackles integrating autonomous agents into distributed healthcare by learning both descriptive norms (medical tendencies) and prescriptive norms (treatment protocols) from real-world data. It introduces MedT and OBJ as a Gaussian-mixture framework for descriptive norms, a SINP-based perception mechanism, and a rational-inductive logic approach within Markov games for prescriptive norms, augmented by external practice verification and adaptive learning. Key contributions include a two-tier norm learning pipeline, EM-based SINP learning with KL-based convergence, and a Bayesian, mean-field prescriptive-norm induction validated on a 2016–2020 vertigo dataset, showing convergence of agent behaviors to current clinical practices. The work advances practical, interaction-driven norm perception and induction in healthcare, enabling safer AI-human collaboration and adaptive policy adherence in distributed clinical settings.

Abstract

This paper presents a Multi-Agent Norm Perception and Induction Learning Model aimed at facilitating the integration of autonomous agent systems into distributed healthcare environments through dynamic interaction processes. The nature of the medical norm system and its sharing channels necessitates distinct approaches for Multi-Agent Systems to learn two types of norms. Building on this foundation, the model enables agents to simultaneously learn descriptive norms, which capture collective tendencies, and prescriptive norms, which dictate ideal behaviors. Through parameterized mixed probability density models and practice-enhanced Markov games, the multi-agent system perceives descriptive norms in dynamic interactions and captures emergent prescriptive norms. We conducted experiments using a dataset from a neurological medical center spanning from 2016 to 2020.

Figures (15)

• Figure 1: The Two-Step Process of Norm Perception in Agents.The model establishes SINP through two steps: (a) Collective information is mapped to prior norm perception. (b) New data is combined with prior perceptions to adapt the mixture parameters, with adjustments based on data amount.
• Figure 2: KL Divergence between $\text{SINP}_i$ and OBJ after practice sharing activities.
• Figure 3: Boxplot showing the distribution of convergence steps for different numbers of agents.
• Figure 4: Histogram of Convergence Steps for 20 Agents over 50 runs.
• Figure 5: Convergence Ratio Decrease with Increasing Number of Agents.