Computational Bayesian Statistics Journal Club
Moderator: Aki Vehtari [Aalto Uni]
Topic: Improvements and limitations of autodiff variational inference (ADVI)
Paper:
– Robust, Accurate Stochastic Optimization for Variational Inference
https://arxiv.org/abs/2009.00666
Additional Papers:
Pareto k diagnostic, difference between pre-asymptotic and asymptotic behavior,
and high dimensional examples are from
– Pareto Smoothed Importance Sampling
https://arxiv.org/abs/1507.02646
Some ideas on why stochastic optimization for other divergences is more difficult comes from:
– Perturbative Black Box Variational Inference
https://arxiv.org/abs/1709.07433
Please email Sara Mejias Gonzalez at [email protected] for Zoom information if you would like to participate.