Output details
11 - Computer Science and Informatics
Keele University
Reservoir computing and extreme learning machines for non-linear time-series data analysis
<24>This work cements a collaboration with the internationally leading reservoir computing (RC) research group at Ghent University. The paper introduces a new RC architecture that uses two random static projections (R2SP). In addition to sharing their fast training time properties, this work explains why R2SP is superior to conventional RC techniques when good fading-short-term memory and non-linear processing are simultaneously required (e.g. an important challenge in speech recognition). Using a challenging benchmark dataset the paper provides a detailed comparative study between R2SP, conventional RC and a third network architecture with a fixed short-term memory in order to demonstrate R2SP’s superiority.