Generative modeling and deep learning theory papers explained
A mathematical view can explain why a learning objective works and where its assumptions enter. This collection studies latent-variable models, adversarial learning, continuous-depth networks, and compression. Each explainer builds the intuition before working through the formal construction.
Compare two ways to learn a generative model, then change perspective from discrete network layers to continuous dynamics. Treat the differences in objectives and assumptions as the thread connecting these papers.
Start here
Auto-Encoding Variational Bayes
Connect latent variables, approximate inference, and the variational training objective.
Generative Adversarial Networks
Compare likelihood-based modeling with a generator trained against a discriminator.
Neural Ordinary Differential Equations
Interpret a network as a differential equation and follow how it can be trained.
All Theory explainers
4 papers · newest explainers first
