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

By Sam Bond-Taylor, Adam Leach, Yang Long, Chris G. Willcocks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

712 citations recorded by the source corpus.

Deep generative models are a class of techniques that train deep neural networks to model the distribution of training samples. Research has fragmented into various interconnected approaches, each of which make trade-offs including run-time, diversity, and architectural restrictions. In particular, this compendium covers energy-based models, variational autoencoders, generative adversarial networks, autoregressive models, normalizing flows, in addition to numerous hybrid approaches. These techniques are compared and contrasted, explaining the premises behind each and how they are interrelated, while reviewing current state-of-the-art advances and implementations.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/bc519f58ae61afbf6318d6e4239d2d565c7ba467

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

RODIN: A Generative Model for Sculpting 3D Digital Avatars Using Diffusion (2022, background): “The state of generative modeling [5,14,15,28,48,65,75] has seen rapid progress in past years.”
Efficient-VDVAE: Less is more (2022, background): “Variational Autoencoder[5, 6] (VAE) is a different type of maximum likelihood models that are usually associated with representation learning [7, 8] and are known to generate subjectively more diverse images[9] compared to other generative approaches like diffusion models [10, 11, 12] and GANs [8, 1”
Chunked Autoregressive GAN for Conditional Waveform Synthesis (2021, background): “These advancements are driven by concurrent improvements in deep generative models (Bond-Taylor et al., 2021), which model the distribution of training data with or without auxiliary conditioning information.”
Pros and Cons of GAN Evaluation Measures: New Developments (2021, background): “They are remarkably eﬀective at synthesizing strikingly realistic and diverse images, and learn to do so in an unsupervised manner (Wang et al., 2019; Liu et al., 2021; Bond-Taylor et al., 2021).”
AnoDDPM: Anomaly Detection with Denoising Diffusion Probabilistic Models using Simplex Noise (2022, background): “Denoising diffusion probabilistic models (DDPMs) [7] are capable of generating samples from complex data distributions, with superior mode coverage [27] over generative adversarial networks (GANs) and variational autoencoders (VAEs) [4] (a desirable property for anomaly detection and small datasets)”
Unleashing the Power of Edge-Cloud Generative AI in Mobile Networks: A Survey of AIGC Services (2023, neutral): “The study in [6] provides comprehensive guidance and comparison among advanced generative models, including GANs, energy-based models, variational autoencoder (VAE), autoregressive models, ﬂow-based models, and diffusion models.”
Diffusion models in medical imaging: A comprehensive survey (2022, methodology): “The energy-based models (EBMs) achieve this by defining an unnormalized probability density over a state space; however, these methods require Markov Chain Monte Carlo (MCMC) sampling during both training and inference, which is a slow iterative process [7].”
A Survey on Generative Diffusion Models (2022, methodology): “These models effectively sur-mount the obstacles entailed in aligning posterior distributions within VAEs, mitigating the instability inherent in adversarial objectives of GANs, addressing the computational burdens associated with Markov Chain Monte Carlo (MCMC) methods during training in EBMs, and…”