Learning a sequence of tasks without access to iid observations is a widely studied form of
continual learning (CL) that remains challenging. In principle, Bayesian learning directly …