Approximate Bayesian Updating: Sequential Inference for Streaming Data

Bayesian models can be updated if additional data is observed, improving the quality of produced forecasts. However, the computational algorithms for Bayesian statistics are generally designed for only one update, and cannot be run again to add more data without discarding the previous result and starting again. This thesis proposes several approximate methods that allow models to be updated repeatedly, for use in applications when data is constantly being observed. These methods are shown to be faster than alternatives without significant loss of accuracy.