The second addresses scalability in simulation-based inference through data reduction, developing methods that learn informative low-dimensional summaries to enable efficient inference in complex settings, with applications to gravitational wave analysis.
Keywords: GBI, SBI, amortized learning
We introduce a Bayesian framework for hyperparameter inference using held-out data, enabling coherent estimation and uncertainty quantification, together with amortized generalized-posterior approximations that avoid repeated costly sampling across datasets and hyperparameter values.
Abstract: My talk consists of two parts. The first focuses on Generalised Bayesian Inference (GBI), where the loss hyperparameters are critical under model misspecification yet difficult to determine in a principled way.
The next OWABI seminar www.warwick.ac.uk/owabi of the Season is tomorrow, Thursday the 28th May at 10am UK time. Kate Lee (University of Auckland) will talk about "Towards Robust and Scalable Bayesian Learning".
teams.microsoft.com/...
Meeting ID: 382 746 871 856 192
Passcode: FZ2gn73K