Continuous-Time Meta-Learning with Forward Mode Differentiation

TL;DR

This paper introduces Continuous-Time Meta-Learning (COMLN), a novel approach that models adaptation as an ODE, enabling efficient, stable, and flexible meta-learning with continuous trajectories and exact gradient computation.

Contribution

The paper proposes COMLN, a continuous-time meta-learning algorithm using forward mode differentiation for exact gradients, allowing longer adaptation with constant memory and improved stability.

Findings

COMLN achieves efficient meta-learning with reduced memory usage.

It provides theoretical guarantees for stability.

Empirical results show effectiveness on few-shot image classification.

Abstract

Drawing inspiration from gradient-based meta-learning methods with infinitely small gradient steps, we introduce Continuous-Time Meta-Learning (COMLN), a meta-learning algorithm where adaptation follows the dynamics of a gradient vector field. Specifically, representations of the inputs are meta-learned such that a task-specific linear classifier is obtained as a solution of an ordinary differential equation (ODE). Treating the learning process as an ODE offers the notable advantage that the length of the trajectory is now continuous, as opposed to a fixed and discrete number of gradient steps. As a consequence, we can optimize the amount of adaptation necessary to solve a new task using stochastic gradient descent, in addition to learning the initial conditions as is standard practice in gradient-based meta-learning. Importantly, in order to compute the exact meta-gradients required for the outer-loop updates, we devise an efficient algorithm based on forward mode differentiation, whose memory requirements do not scale with the length of the learning trajectory, thus allowing longer adaptation in constant memory. We provide analytical guarantees for the stability of COMLN, we show empirically its efficiency in terms of runtime and memory usage, and we illustrate its effectiveness on a range of few-shot image classification problems.