Learning to Sample Tasks for Meta Learning
Lists
Only you see this
Abstract
Through experiments on various meta-learning methods, task samplers, and few-shot learning tasks, this paper arrives at three conclusions. Firstly, there are no universal task sampling strategies to guarantee the performance of meta-learning models. Secondly, task diversity can cause the models to either underfit or overfit during training. Lastly, the generalization performance of the models are influenced by task divergence, task entropy, and task difficulty. In response to these findings, we propose a novel task sampler called Adaptive Sampler (ASr). ASr is a plug-and-play task sampler that takes task divergence, task entropy, and task difficulty to sample tasks. To optimize ASr, we rethink and propose a simple and general meta-learning algorithm. Finally, a large number of empirical experiments demonstrate the effectiveness of the proposed ASr.
BibTeX
@article{wang2024learning,
author = {Jingyao Wang and Zeen Song and Xingzhe Su and Lingyu Si and Hongwei Dong and Wenwen Qiang and Changwen Zheng},
title = {{Learning to Sample Tasks for Meta Learning}},
journal = {Int. J. Comput. Vis.},
year = {2024},
url = {https://openreview.net/forum?id=VcJrySO71d},
}| The Effect of Diversity in Meta-Learning | arXiv · 2022 |
| Adaptive Task Sampling for Meta-learning | arXiv · 2020 |
| Meta-learning with Heterogeneous Tasks | arXiv · 2024 |
| Adversarial Task Up-sampling for Meta-learning | arXiv · 2022 |
| Task-Agnostic Meta-Learning for Few-shot Learning | CVPR · 2019 |
| Meta-learning with an Adaptive Task Scheduler | NeurIPS · 2021 |