Variational Auto-Regressive Gaussian Processes for Continual Learning

                Sanyam Kapoor 1 Theofanis Karaletsos 2 Thang D. Bui 3

             Abstract
   Through sequential construction of posteriors on
   observing data online, Bayes’ theorem provides
   a natural framework for continual learning. We
   develop Variational Auto-Regressive Gaussian
   Processes (VAR-GPs), a principled posterior up-
   dating mechanism to solve sequential tasks in
   continual learning. By relying on ...