arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2301.11779cs.LGcs.CV
收藏

面向分布外泛化的不变元学习

Invariant Meta Learning for Out-of-Distribution Generalization

The Australian National University(澳大利亚国立大学)

Penghao Jiang, Ke Xin, Zifeng Wang, Chunxi Li

摘要:

现代深度学习技术已经在许多领域展示了其出色的能力,但依赖于大量的训练数据。基于优化的元学习训练模型是针对多种任务进行的,因此它可以使用少量的训练样本来解决新的学习任务,但是这些方法都假设训练数据和测试数据是相同且独立分布的。为了克服这种局限性,在本文中,我们提出了不变量元学习的分布任务。具体来说,不变元学习找到不变的最优元初始化,并快速适应正则化惩罚的分布外任务。大量的实验表明,我们提出的不变元学习的有效性外的分布Few-Shot任务。

英文摘要:

Modern deep learning techniques have illustrated their excellent capabilities in many areas, but relies on large training data. Optimization-based meta-learning train a model on a variety tasks, such that it can solve new learning tasks using only a small number of training samples.However, these methods assumes that training and test dataare identically and independently distributed. To overcome such limitation, in this paper, we propose invariant meta learning for out-of-distribution tasks. Specifically, invariant meta learning find invariant optimal meta-initialization,and fast adapt to out-of-distribution tasks with regularization penalty. Extensive experiments demonstrate the effectiveness of our proposed invariant meta learning on out-of-distribution few-shot tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑