A Distribution-Dependent Analysis of Meta-Learning

                Mikhail Konobeev 1 Ilja Kuzborskij 2 Csaba Szepesvari 1 2

             Abstract               “similar” to the previous ones. In the formal framework
   A key problem in the theory of meta-learning       of statistical meta-learning of Baxter (2000), the learner is
   is to understand how the task distributions influ-    given a sequence of training “sets”. The data in each set is
   ence transfer risk, the expected error of a ...