Statistically and Computationally Efficient Linear Meta-representation Learning

Neural Information Processing Systems 

In typical few-shot learning, each task is not equipped with enough data to be learned in isolation.