Gaussian Processes

The Gaussian process as a tool for, predominantly, regression tasks in machine learning has only been growing in popularity over recent years. Although not reaching the same widespread usage as neural network-based technology, it is also considered a key methodology for the machine learning pratictioner. In this short chapter, a basic introduction to the approach will be provided; following which, several extensions to the fundamental Gaussian process will be presented; finally, a case study involving wind turbine power curves will serve to highlight the engineering benefits of this approach. Two particular strengths will be brought to the fore: firstly, the automated quantification of uncertainty (under a Bayesian paradigm) inherent to the Gaussian process approach; secondly, the flexibility of the model compared to the relatively small number of tunable hyperparameters to be learnt.