Tag
This paper analyzes first-order meta-learning algorithms for few-shot learning, introducing Reptile and providing theoretical insights into why these computationally efficient methods work well on established benchmarks.
OpenAI introduces Reptile, a scalable meta-learning algorithm for few-shot classification that achieves comparable performance to MAML while converging faster with lower variance. The paper provides theoretical analysis showing Reptile maximizes inner product between task gradients for improved generalization.
OpenAI researchers introduce E-MAML and E-RL², two meta-reinforcement learning algorithms designed to improve exploration in tasks where discovering optimal policies requires significant exploration. The work demonstrates these algorithms' effectiveness on novel environments including Krazy World and maze tasks.