Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks

TLDR

Meta‑learning trains models across many tasks so they can solve new tasks with few samples. The authors propose a model‑agnostic meta‑learning algorithm that works with any gradient‑descent‑trained model for classification, regression, and reinforcement learning. The algorithm trains model parameters so that a few gradient steps on limited data from a new task yield strong generalization. The method achieves state‑of‑the‑art few‑shot classification, strong few‑shot regression performance, and faster fine‑tuning for policy‑gradient reinforcement learning.

Abstract

We propose an algorithm for meta-learning that is model-agnostic, in the sense that it is compatible with any model trained with gradient descent and applicable to a variety of different learning problems, including classification, regression, and reinforcement learning. The goal of meta-learning is to train a model on a variety of learning tasks, such that it can solve new learning tasks using only a small number of training samples. In our approach, the parameters of the model are explicitly trained such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance on that task. In effect, our method trains the model to be easy to fine-tune. We demonstrate that this approach leads to state-of-the-art performance on two few-shot image classification benchmarks, produces good results on few-shot regression, and accelerates fine-tuning for policy gradient reinforcement learning with neural network policies.