An Improved Eulerian Echo State Network for Static Temporal Graphs

Machine learning for graphs has gained significant attention in recent years, with a focus on developing efficient and effective approaches. Traditional methods often rely on complex deep neural networks and computationally demanding training algorithms, highlighting the need for alternative solutions. Reservoir Computing (RC) models might be crucial in this situation since they make it possible to create useful graph embeddings using untrained recursive structures. In this paper, we propose the use of an antisymmetric matrix and an Eulerian Graph Echo State Network (EuGESN) for graph-based machine learning. Antisymmetric matrices are employed to initialize the reservoir of the Graph Echo State Network (GESN), which helps capture the dynamics and relationships within the graph structure. Our approach has resulted in promising results on six different datasets.