RotoGBML: Towards Out-of-Distribution Generalization for Gradient-Based Meta-Learning

TL;DR

RotoGBML introduces a method to improve gradient-based meta-learning's out-of-distribution generalization by homogenizing task gradients through reweighting and rotation, focusing on invariant causal features for better adaptation.

Contribution

The paper proposes RotoGBML, a novel approach that homogenizes OOD task gradients using reweighted vectors and rotation matrices, enhancing meta-learning robustness.

Findings

Outperforms state-of-the-art methods on few-shot image classification benchmarks.

Effectively homogenizes task gradients across diverse distributions.

Focuses on invariant causal features to improve generalization.

Abstract

Gradient-based meta-learning (GBML) algorithms are able to fast adapt to new tasks by transferring the learned meta-knowledge, while assuming that all tasks come from the same distribution (in-distribution, ID). However, in the real world, they often suffer from an out-of-distribution (OOD) generalization problem, where tasks come from different distributions. OOD exacerbates inconsistencies in magnitudes and directions of task gradients, which brings challenges for GBML to optimize the meta-knowledge by minimizing the sum of task gradients in each minibatch. To address this problem, we propose RotoGBML, a novel approach to homogenize OOD task gradients. RotoGBML uses reweighted vectors to dynamically balance diverse magnitudes to a common scale and uses rotation matrixes to rotate conflicting directions close to each other. To reduce overhead, we homogenize gradients with the features rather than the network parameters. On this basis, to avoid the intervention of non-causal features (e.g., backgrounds), we also propose an invariant self-information (ISI) module to extract invariant causal features (e.g., the outlines of objects). Finally, task gradients are homogenized based on these invariant causal features. Experiments show that RotoGBML outperforms other state-of-the-art methods on various few-shot image classification benchmarks.