La-MAML: Look-ahead Meta Learning for Continual Learning

Neural Information Processing Systems 

The continual learning problem involves training models with limited capacity to perform well on a set of an unknown number of sequentially arriving tasks.