PROJECT TITLE :

Deep Generative Modelling A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models

ABSTRACT:

Deep generative models are a category of methods that train deep neural networks to model the distribution of training samples. These models can be used to generate new data from existing data. The field of research has splintered into many different interconnected approaches, each of which has its own set of compromises, such as those regarding run-time, diversity, and architectural restrictions. This compendium specifically covers energy-based models, variational autoencoders, generative adversarial networks, autoregressive models, and normalizing flows, in addition to a wide variety of hybrid methodologies. These methods are contrasted and compared, with an explanation of the premises behind each one and how they are interrelated, followed by a discussion of the most recent advances in state-of-the-art implementations and developments.