Wednesday 09 April 2025
A new approach to predicting coalition formation in complex systems has been developed, using a deep learning model to estimate power indices for voting games. The researchers behind this work have created a novel neural network architecture that efficiently estimates these power indices, which are essential in assessing the contribution and influence of individual agents in multi-agent systems.
The study focuses on cooperative game theory, where multiple agents collaborate to achieve a common goal. In these situations, understanding how each agent’s contributions affect the outcome is crucial. The Shapley value and Banzhaf index are two popular methods used to measure an agent’s power and influence in coalition formation. However, these traditional approaches can be computationally expensive and impractical for large-scale systems.
The new neural network model, dubbed InfluenceNet, addresses this limitation by providing a more efficient way to estimate power indices. The architecture is designed to learn from datasets generated using various randomization methods, including uniform and Gaussian distributions. By training the model on these datasets, it can learn patterns in coalition formation and predict the outcome of different scenarios.
The researchers tested InfluenceNet on three types of datasets: coin-flip random, Mixture of Gaussian (MoG), and uniformly distributed random. The results show that the model performs well when evaluated on conditions it was trained on, but struggles with learning the different distributions of the datasets themselves. This is particularly evident in the MoG dataset, where the model’s performance deteriorates significantly.
The study highlights the importance of considering the number of agents and the distribution of rule values when evaluating coalition formation models. The researchers found that as the distance between the training and testing datasets increases, the model’s performance decreases. This underscores the need for more advanced neural network architectures that can handle dynamic systems better.
InfluenceNet has potential applications in various fields, including economics, politics, and computer science. It could be used to analyze complex decision-making processes, identify influential agents, and predict outcomes of coalition formation. The study demonstrates a promising step towards developing more efficient and effective methods for understanding coalition formation in multi-agent systems.
The researchers plan to further improve the model by incorporating additional features and exploring new neural network architectures. As the field continues to evolve, InfluenceNet could play a significant role in advancing our understanding of complex systems and decision-making processes.
Cite this article: “Unlocking the Power of Cooperative Games: A Neural Network Approach to Predicting Fair and Stable Payoff Allocations”, The Science Archive, 2025.
Coalition Formation, Deep Learning, Power Indices, Voting Games, Cooperative Game Theory, Multi-Agent Systems, Influencenet, Neural Network Architecture, Shapley Value, Banzhaf Index
