Variational Inference
1 like686 views
The document discusses scalable approximate inference methods, focusing on variational inference and sampling techniques for complex probabilistic models in recommender systems. It emphasizes the Bayesian mixture model and the application of KL-divergence and the evidence lower bound (ELBO) for optimization. The content also covers choosing variational distributions and mean-field structured variational inference.
Variational Inference
- 1. Part 2: Scalable Approximate Inference Session 1: - Variational Inference Session 2: - Sampling methods Approximate and Scalable Inference for Complex Probabilistic Models in Recommender Systems
- 2. Introduction
- 3. Motivation: bayesian mixture model
- 4. Main idea
- 5. KL-Divergence KL of the posterior
- 7. ELBO e KL-Divergence Jensen inequality concave (log) KL of the posterior Evidence lower bound (ELBO) + +
- 8. KL and ELBO Choose family of variational distributions such that the expectations of log(q(z)) and log(p(x,z)) are computable
- 9. Mean-field
- 10. Optimizing a functional Euler-lagrange equation
- 11. Mean-field