Manipulation of Machine Learning Algoirhtms

As data becomes increasingly available, individuals, organisations and companies are increasingly applying machine learning algorithms to make decisions. In many cases, those decisions have a direct effect on those who provided the data to the decision maker. In other words, data providers often have a vested interest in the decisions made based on the data provided. Therefore, decision makers should anticipate that data providers may alter or change the data they provide in order to achieve a preferential outcome. Such strategic behaviour is not adequately modelled by classical machine learning settings in the literature. As a result, new machine learning algorithms are required, which take into the account the incentives and capabilities of data providers when making decisions. This paper summarises a PhD project which attempts to address this problem in a number of contexts.