A Comparative Analysis of Reservoir Computing Methods : A Comparison of Liquid State Machines and Echo State Networks in Reservoir Computing

Abstract

Reservoir Computing (RC) has emerged as a powerful paradigm in machine learning, offering simplicity, efficiency, and impressive performance across diverse tasks. However, the choice of reservoir architecture can significantly impact performance and computational demands. This thesis investigates two prominent RC architectures: Liquid State Machines (LSMs) and Echo State Networks (ESNs), and aims to address the challenge of selecting the most effective and efficient RC architecture for specific applications. We compare the performance and resource requirements of LSMs and ESNs, focusing on their suitability for real-world tasks with imbalanced data, such as epileptic seizure detection. The findings suggest that LSMs can achieve significantly higher accuracy compared to ESNs. However, this potential performance gain comes at the cost of increased computational demands. The research also indicates that LSMs could be particularly beneficial for real-world tasks like seizure detection due to their potential robustness to imbalanced data.