Discussion Lead: Bob Carpenter (CCM)
Topic: (1) Advances in Black-Box VI: Normalizing Flows, Importance Weighting, and Optimization Abhinav Agrawal, Daniel Sheldon, Justin Domke. 2020.
(2) Robust, Automated, and Accurate Black-box Variational Inference. Manushi Welandawe, Michael Riis Andersen, Aki Vehtari, Jonathan H. Huggins. 2022.
Abstract: We’re taking a second look at these papers because Justin’s here visiting and we’re going to be building approach (1) below out to fit all the models in posteriordb using BridgeStan. The papers come to rather different conclusions on the utility of normalising flows, and I’m guessing this is likely due to secondary factors around how they were optimized.
Paper Link (1): https://arxiv.org/abs/2006.10343
Paper Link (2): https://arxiv.org/abs/2203.15945