Distributed SVMs Using Diffusion Strategy

Ensemble learning combines multiple weak learners with the expectation of obtaining a better model. How multiple learners work together and interact is critical to effectiveness of model. In this paper, we propose a Multi-agent SVMs method based on distributed networks to improve the performance of SVM methods. By forming different data subsets for each SVM and applying a Bagging mechanism to the results, we improve the graph network's generalization performance. In addition, the Diffusion strategy is used in the training phase of the network, enabling each agent to improve its classification performance. Through experiment comparison, our method achieves a leading grade on public datasets.