probml / pml2-book

Probabilistic Machine Learning: Advanced Topics
MIT License
1.39k stars 120 forks source link

The bound in VI is concave (and other typos/errors if found) #257

Closed maremita closed 1 year ago

maremita commented 1 year ago

[Version: 2023-04-01]

In page 451, just before Algorithm 10.4:

Convergence is guaranteed since the bound is convex wrt each of the factors $q_i$ [Bis06, p. 466].

The bound (ELBO) should be concave wrt each of the factor $q_i$.

Thanks.

maremita commented 1 year ago

[Version: 2023-04-01]

In the bibliography, reference : [Gre98] P. Green. “ Reversible Jump Markov Chain Monte Carlo computation and Bayesian model determination”. In: Biometrika 82 (1998), pp. 711–732. The year of publication should be 1995.

Link to the paper

Thanks.

murphyk commented 1 year ago

fixed