Task Agnostic Continual Learning Using Online Variational Bayes with Fixed-Point Updates