Refined probabilistic abstraction

Computer networks and embedded systems are ubiquitous and critical parts of our daily life. Therefore performance and reliability guarantees for these systems are crucial. To this end, versatile probabilistic modelling and analysis techniques have been developed. However existing probabilistic analysis methods are inherently limited to small systems. This dissertation introduces a new probabilistic analysis method that scales to large and even infinite systems which are far out of reach of previous methods. The key idea is to approximate a given system by a smaller abstraction which is refined automatically until sufficient precision has been achieved. The thesis discusses the various foundational and practical challenges involved in developing this method, as well as its effectiveness in practice.