Graph Centrality Pruning Boosts Echo State Network Efficiency
Researchers propose a graph centrality-based pruning method for Echo State Networks (ESNs) that removes redundant reservoir nodes. By interpreting the reservoir as a weighted directed graph and applying centrality measures, the method reduces computational overhead while maintaining or improving prediction accuracy. Experiments on Mackey-Glass time-series and electric load forecasting confirm the approach's effectiveness. The work is published on arXiv (2603.20684) under Computer Science > Machine Learning.
Key facts
- Echo State Networks are used for nonlinear time-series prediction.
- Randomly initialized reservoirs often contain redundant nodes.
- The proposed method uses graph centrality measures for pruning.
- Reservoir is interpreted as a weighted directed graph.
- Experiments on Mackey-Glass time-series prediction.
- Experiments on electric load forecasting.
- Method reduces reservoir size while maintaining or improving accuracy.
- Published on arXiv with ID 2603.20684.
Entities
Institutions
- arXiv