Is Meta-Learning Out? Rethinking Unsupervised Few-Shot Classification with Limited Entropy
This work addresses the problem of evaluating and enhancing meta-learning for unsupervised few-shot classification, providing insights for researchers in machine learning, though it is incremental as it builds on existing comparisons and methods.
The paper challenges the necessity of meta-learning for few-shot classification by showing that whole-class training can match its performance, but demonstrates through theoretical and experimental analysis that meta-learning offers tighter generalization bounds and greater robustness to label noise and heterogeneous tasks, leading to the proposal of MINO, a meta-learning framework that improves unsupervised performance in few-shot and zero-shot tasks.
Meta-learning is a powerful paradigm for tackling few-shot tasks. However, recent studies indicate that models trained with the whole-class training strategy can achieve comparable performance to those trained with meta-learning in few-shot classification tasks. To demonstrate the value of meta-learning, we establish an entropy-limited supervised setting for fair comparisons. Through both theoretical analysis and experimental validation, we establish that meta-learning has a tighter generalization bound compared to whole-class training. We unravel that meta-learning is more efficient with limited entropy and is more robust to label noise and heterogeneous tasks, making it well-suited for unsupervised tasks. Based on these insights, We propose MINO, a meta-learning framework designed to enhance unsupervised performance. MINO utilizes the adaptive clustering algorithm DBSCAN with a dynamic head for unsupervised task construction and a stability-based meta-scaler for robustness against label noise. Extensive experiments confirm its effectiveness in multiple unsupervised few-shot and zero-shot tasks.