/Probabilistic-Deep-Learning

Probabilistic Deep Learning

This course contains the lecture notes of the course "Probabilistic deep learning" given in the Master's programme in AI of Radboud university. The material is at a preliminal stage and it likely contains typos, small errors and inaccuracies.

Provisional table of content

Introduction

  1. Probabilistic models and maximum likelihood
  2. Deep learning

Part 1: Variational inference

  1. Univariate Bayesian inference
  2. Univariate Bayesian inference by gradient descent
  3. Multivariate inference
  4. From importance sampling to stochastic variational inference
  5. Timeseries analysis and structured variational inference
  6. Amortized inference and inference networks
  7. Simultaneous variational inference and maximum likelihood
  8. Variational gradient descent for discrete models

Part 3: Advanced Variational inference

  1. Gradient estimators
  2. Variance reduction
  3. Variational inference with normalizing flows
  4. Black-box and likelihood-free inference
  5. Probabilistic programming and automatic inference

Part 2: Variational methods for machine learning

  1. Variational supervised learning
  2. Bayesian neural networks
  3. Generative modeling and variational autoencoders
  4. Ensamble methods
  5. Probabilistic metalearning
  6. Probabilistic geometric deep learning

Part 3: Inference and control

  1. Bayesian decision theory
  2. KL control theory and reinforcment learning
  3. Belief-augmented reinforcement learning