Discussed the following papers:
-
Blei, D. M., Kucukelbir, A., and McAuliffe, J. D. (2017). Variational inference: a review for statis- ticians. Journal of the American Statistical Associ- ation, 112:855–877.
-
Huggins, J. H., Kasprzak, M., Campbell, T., and Broderick, T. (2019). Practical posterior error bounds from variational objectives. arXiv preprint arXiv:1910.04102.
-
Yao, Y., Vehtari, A., Simpson, D., and Gelman, A. (2018). Yes, but did it work?: Evaluating variational inference. arXiv preprint arXiv:1802.02538.
-
Vehtari, A., Gelman, A., and Gabry, J. (2015). Pareto smoothed importance sampling. arXiv preprint arXiv:1507.02646.
... and many others.
Ended up comparing the various VI frameworks (Pyro, PyMC3, Theano, ...) as many were not flexible enough for different application problems.