Risk-Averse Decision-Making under Parametric Uncertainty

Licentiate thesis, 2022

For sequential decision-making problems with potentially catastrophic consequences appropriate risk assessment may be required. In contrast to traditional techniques for decision-making under uncertainty that aim to maximise performance in expectation, we chose to focus on other properties of the probability distribution. For instance, in an application such as autonomous driving, the chance of causing an accident might be small but yet fatal. A decision-maker focusing on performance in the worst outcomes may be able to obtain a safer decision-making process by keeping this in mind. We propose frameworks for quantifying uncertainty under the reinforcement learning framework and develop algorithms that allow for risk-sensitive decision-making under uncertainty.