The Impacts of Diffusion Kernels on Recognition-Primed Multi-agent Decision Making

Novel perspectives on multiscale information analysis are highly demanded in several areas of multi-agent research, including large-scale agent organization and experience-based decision making. A recent breakthrough in harmonic analysis is diffusion geometry and diffusion wavelets, which offers a general framework for multiscale analysis of massive data sets. In this paper, we investigate the impacts of various diffusion kernel functions on the performance of solution synthesis in experience-based multi-agent decision making. In particular, we take a two-phase approach to conduct our experiment. In phase one (learning), we apply four commonly-used kernel functions on a data set including a large collection of battlefield decision experiences, identifying the kernel functions that can outperform the others. In phase two, we apply the kernels identified in phase one to an independent data set for testing. It is shown that Cosine exponential outperformed the other kernel functions. In general, this study indicates that, in order to achieve the best possible performance in diffusion multiscale analysis, it is critical to identify kernel functions that are applicable to the massive data set under concern.