RML4FES

Machine learning algorithms have demonstrated performance near human-level in many real-world applications, resulting in replacing human supervision. For example, in recent years, machine learning algorithms such as deep learning have been successfully deployed in a real-life setting, such as automated decision making in smart-energy systems. Despite these significant contributions, the robustness of most existing machine learning algorithms is uncertain. Recently there is a growing interest in more robust and generalisable algorithms in a machine learning community. A critical aspect of a robust machine learning algorithm is the learning algorithms with the ability to quantify the predictive uncertainty and provide generalised performance in low data settings.

Thus, this research project aims to:

1. Investigate and develop robust machine learning algorithms that provide a reasonable estimate of predictive uncertainty
2. Develop robust machine learning algorithms capable of learning in a low data setting (the presence of limited labelled training data and worst-case noise such as outliers and missing values.
3. Assess the applicability of the designed methods in real-world energy system data.