Masked Autoregressive Flows with a Marginalising Flow Framework
Invertible deep networks, or normalising flows, are the focus of an influx of new research in the field of machine learning. In this paper marginalising flows, which can be applied on top any normalising flow, will be implemented over the existing Masked Autoregressive Flow framework. The performance will be compared between different masked autoregressive models with and without marginalising flows. Additionally image generation will be analysed after training on CIFAR-10.
Faculteit der Sociale Wetenschappen