In the field of continual learning, the objective is to learn several tasks
one after the other without access to the data from previous tasks. Several
solutions have been proposed to tackle this problem but they usually assume
that the user knows which of the tasks to perform at test