TL
TL;DR
A family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates, including Reptile, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task.
Abstract
This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously unseen task sampled from this distribution. We analyze a family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates. This family includes and generalizes first-order MAML, an approximation to MAML obtained by ignoring second-order derivatives. It also includes Reptile, a new algorithm that we introduce here, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task. We expand on the results from Finn et al. showing that first-order meta-learning algorithms perform well on some well-established benchmarks for few-shot classification, and we provide theoretical analysis aimed at understanding why these algorithms work.
References22 items
1
Recasting Gradient-Based Meta-Learning as Hierarchical Bayes
2
Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm
3
Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions
4
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
5