Deep Generative Model

DeepLearningMachineLearningNeuralNetworksProbabilityTheoryStatistics A deep generative model is one that utilizes a deep neural network to explicitly or implicitly learn a Generative Model. The chain rule of conditional probability gives an exponential number of dependencies between random variables in modeling a full joint distribution while a Bayesian Network assumes conditional independencies to simply the above factorization, a deep generative model assumes the conditional distributions can be modeled using neural networks: The following categories coarsely divide the different types of deep generative models:

• Models with full likelihood
  • Autoregressive Models
  • Normalizing Flows
• Latent variables models:
  • Variational inference
  • VAEs
• Implicit models (model sampling procedure)
  • Embeddings:
  • F-divergences:
  • GANs
• EBMs
• Diffusion models