Survey of Likelihood-based Generative Models Keyword: VAE, Normalizing Flows, Neural ODE, Energy-based Models, Diffusion Models, Score Models, Schrodinger Bridge, Rectified Flows, Flow Models, Consistency Models, Flow Map Models, Distribution Matching Distillation, Drifting Models Abstract

Hyunjik kim et al., 2019, arXiv Keyword: Bayesian, Process Problem: Underfitting of Neural Process Solution: NP + Self-Attention, Cross-Attention Benefits: Improvement of prediction accuracy, training speed, model capability.

Marta Garnelo et al., 2018, arXiv Keyword: Bayesian, Process Problem: Data inefficiency, hard to train multiple datasets in one.

Marta Garnelo et al., 2018, arXiv Keyword: Bayesian, Process Problem: Weakness of knowledge sharing and data inefficiency of classical supervised learning Solution: Stochastic Process + NN Benefits: Data efficient, prior sharing Contribution: Encapsulation of parameterized NN function family.