Game theory is known to be a useful tool in the study of Machine Learning (ML) and Artificial Intelligence (AI) Multi-Agent interactions. One basic component of these ML and AI systems is the exploration-exploitation trade-off, a fundamental dilemma between taking a risk with new actions in the quest for more information about the environment (exploration) and repeatedly selecting actions that result in the current maximum reward or (exploitation). However, the outcome of the exploration-exploitation process is often unpredictable in practice and the reasons behind its volatile performance have been a long-standing open question in the ML and AI communities. Dr Stefanos Leonardos and Assistant Professor Georgios Piliouras, researchers from the Singapore University of Technology and Design (SUTD), applied analytical tools from the theory of dynamical systems in the study of multi-agent systems and established a deep connection between exploration-exploitation and Catastrophe Theory (Figures 1 and 2). The latter is a branch of mathematics that formally explains phase transitions in all kinds of natural systems ranging from the transition from water to ice and disease outbreaks to collapses of financial markets.

Game theory is known to be a useful tool in the study of Machine Learning (ML) and Artificial Intelligence (AI) Multi-Agent interactions.