Reservoir Computing is a paradigm that has emerged during the last decade as viable model of how generic neural circuits can be applied to a range of classification and regression machine learning tasks.  Shown to deal elegantly with a range of real-world spatio-temporal signals such as speech and human motion, Reservoir Computing has demonstrated an advantage over more orthodox neural network techniques.

In this talk, we reveal our work on extending a Reservoir Computing model that can learn long-term, non-stationary data -- 50 years of weather data in our case.  To allow the neural circuit to adapt to seasonal and climatic changes over the time period, we apply a range of regulated Hebbian based plasticity rules to the neural synapses, as observed in real, biological neural networks from neuroscience.