Modelling Data with Stochastic Generative Processes

Learning and modelling complex data with generative models is a prominent research topic in modern artificial intelligence since generative models empower us to answer fundamental and important research questions into the generative processes of algorithmic intelligence. They also enable practical applications such as recognising patterns and anomaly, constructing high-level abstractions utilised in reasoning and decision making, and drawing deep insights from the data to facilitate data compression and augmentation. This dissertation aims to advance research in generative models by enriching their modelling capacity and enhancing their robustness to work more efficiently on a wider range of problems and data types.