Bayesian inference for the learning rate in Generalised Bayesian inference

Abstract: In Generalised Bayesian Inference (GBI), the learning rate and hyperparameters of the loss must be estimated. However, these inference-hyperparameters can't be estimated jointly with the other parameters by giving them a prior, as we discuss. Several methods for estimating the learning rate have been given which elicit and minimise a loss based on the goals of the overall inference (in our case, prediction of new data). However, in some settings there exists an unknown true'' learning rate about which it is meaningful to have prior belief and it is then possible to use Bayesian inference with held out data to get a posterior for the learning rate. We give conditions under which this posterior concentrates on the optimal rate and suggest hyperparameter estimators derived from this posterior. The new framework supports joint estimation and uncertainty quatification for inference hyperparameters. Experiments show that the resulting GBI-posteriors out-perform Bayesian inference on simulated test data and select optimal or near optimal hyperparameter values in a large real problem of text analysis. Generalised Bayesian inference is particularly useful for combining multiple data sets and most of our examples belong to that setting. As a side note we give asymptotic results for some of the specialmulti-modular'' Generalised Bayes posteriors, which we use in our examples.