Hacking Task Confounder in Meta-Learning
This addresses a fundamental issue in meta-learning that can degrade generalization performance, offering a causal solution for improved task adaptation.
The paper tackles the problem of negative knowledge transfer in meta-learning by identifying spurious correlations between task-specific causal factors and labels, which they term 'Task Confounders'. It proposes MetaCRL, a plug-and-play method that eliminates these confounders, achieving state-of-the-art performance on various benchmark datasets.
Meta-learning enables rapid generalization to new tasks by learning knowledge from various tasks. It is intuitively assumed that as the training progresses, a model will acquire richer knowledge, leading to better generalization performance. However, our experiments reveal an unexpected result: there is negative knowledge transfer between tasks, affecting generalization performance. To explain this phenomenon, we conduct Structural Causal Models (SCMs) for causal analysis. Our investigation uncovers the presence of spurious correlations between task-specific causal factors and labels in meta-learning. Furthermore, the confounding factors differ across different batches. We refer to these confounding factors as "Task Confounders". Based on these findings, we propose a plug-and-play Meta-learning Causal Representation Learner (MetaCRL) to eliminate task confounders. It encodes decoupled generating factors from multiple tasks and utilizes an invariant-based bi-level optimization mechanism to ensure their causality for meta-learning. Extensive experiments on various benchmark datasets demonstrate that our work achieves state-of-the-art (SOTA) performance.