Dear all,
Our next speaker is Paul
Bürkner (TU Dortmund University), who will talk about "Amortized Mixture and Multilevel Models" with an abstract reported below.
Abstract: Probabilistic mixture and multilevel models are central building blocks in Bayesian data analysis. However, they remain challenging to estimate and evaluate, especially when the involved likelihoods or priors are analytically intractable. Recent
developments in generative deep learning and simulation-based inference have shown promising results in scaling up Bayesian inference through amortization. Against this background, we have developed specialized neural inference frameworks for estimating Bayesian
mixture and multilevel models. The involved neural architectures are closely mirroring the probabilistic symmetries and conditional (in-)dependencies assumed by these models. This not only speeds up neural network training, but also enables amortized inference
for new datasets of varying number of groups and sample sizes.
Keywords: Amortized Bayesian Inference; Neural Posterior Estimation; Probabilistic Factorization
The MS Teams link to join Paul Bürkner's talk is
Meeting ID: 361 477 079 820
Passcode: C49gr9Gx
Lastly, if you missed Jeremias Knoblauch's seminar, you could now watch the recorded talk on the
OWABI webpage.
I'm looking forward to seeing you in a few weeks,
best,
Massimiliano on the behalf of the OWABI Seminar Organisers
------
Dr. Massimiliano Tamborrino
Reader (Associate Professor) and WIHEA Fellow
Department of Statistics
University of Warwick