Reservoir Computing and Echo State Networks for Time-Series Intelligence

Project Overview

This project focuses on reservoir computing implemented using models such as Echo State Networks (ESNs) and Liquid State Machines (LSMs). In this project, we will develop, optimize, and benchmark reservoir architectures for time-series prediction, signal classification, and dynamical system modeling.

The project emphasizes algorithm design, numerical stability, hyperparameter optimization, and a detailed comparison with conventional machine learning methods such as recurrent neural networks (RNNs) and transformers.

Research Context & Motivation

While physical reservoirs explore computation in matter, software-based reservoir computing remains a powerful and efficient framework for understanding the principles of temporal learning. This topic provides a strong computational counterpart to physical and bio-based projects, enabling rigorous benchmarking, rapid prototyping, and advanced theoretical analysis.

By leveraging the fixed, non-linear dynamics of a high-dimensional reservoir, this research aims to unlock highly efficient training mechanisms where only the readout layer requires optimization. The outcomes of this project will directly contribute to low-latency, low-power intelligence frameworks for complex temporal data processing.

Application & Course Details

Interested candidates should review the formal course details below for their application:

Supervisory Team

• Dr. Raphael Fortulan (Primary Supervisor)
  Email: r.vicentefortulan@hud.ac.uk
• Dr. Clay Palmeira Email: c.palmeira@hud.ac.uk