Model-based Data Markets: A Multi-Broker Game Theoretic Approach

The application of machine learning (ML) in data analytics has become widespread across various fields, including healthcare, finance, marketing, and many others. Current research has extensively explored ML model markets, which are typically coordinated by a single broker acting as an intermediary between data providers and model buyers. However, this solution poses significant challenges due to the monopolistic control it grants the broker. Therefore, we propose a multi-broker market model to overcome these issues. The decision-making processes among multiple brokers are formulated as a multi-stage Stackelberg game designed to maximize the profits of each broker. Subsequently, the existence and uniqueness of nash Equilibrium for brokers’ strategies are proven. We conduct simulation experiments using four types of cost functions for two brokers. The simulation results demonstrate that brokers can reach a Nash Equilibrium through the proposed market model.