arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1807.00516cs.LGstat.ML
收藏

迁移学习的均衡分布适应

Balanced Distribution Adaptation for Transfer Learning

Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所);University of Chinese Academy of Sciences(中国科学院大学);Institute of High Performance Computing, A*STAR(新加坡科技研究局高性能计算研究所);Nanyang Technological University(南洋理工大学)

Jindong Wang, Yiqiang Chen, Shuji Hao, Wenjie Feng, Zhiqi Shen

摘要:

迁移学习通过利用来自源领域的知识来注释具有很少或没有标签的目标领域,取得了令人鼓舞的结果。现有的方法通常寻求最小化域之间的分布差异,例如边缘分布、条件分布或两者。然而,这两个距离往往是平等地对待在现有的算法,这将导致在实际应用中的性能较差。此外,现有的方法通常假设数据集是平衡的,这也限制了它们在实际问题中非常常见的不平衡任务上的性能。为了解决分布自适应问题,本文提出了一种新的迁移学习方法--平衡分布\underline{A}自适应(BDA),它可以自适应地利用边缘和条件分布差异的重要性,并且现有的几种方法可以被视为BDA的特殊情况。在BDA算法的基础上,我们还提出了一种新的加权平衡分布自适应算法(W-BDA)来解决迁移学习中的类不平衡问题。W-BDA不仅考虑了域间分布的自适应性,而且自适应地改变了每个类的权重。为了评估所提出的方法,我们进行了广泛的实验上的几个迁移学习任务,这证明了我们提出的算法的有效性超过几个国家的最先进的方法。

英文摘要:

Transfer learning has achieved promising results by leveraging knowledge from the source domain to annotate the target domain which has few or none labels. Existing methods often seek to minimize the distribution divergence between domains, such as the marginal distribution, the conditional distribution or both. However, these two distances are often treated equally in existing algorithms, which will result in poor performance in real applications. Moreover, existing methods usually assume that the dataset is balanced, which also limits their performances on imbalanced tasks that are quite common in real problems. To tackle the distribution adaptation problem, in this paper, we propose a novel transfer learning approach, named as Balanced Distribution \underline{A}daptation~(BDA), which can adaptively leverage the importance of the marginal and conditional distribution discrepancies, and several existing methods can be treated as special cases of BDA. Based on BDA, we also propose a novel Weighted Balanced Distribution Adaptation~(W-BDA) algorithm to tackle the class imbalance issue in transfer learning. W-BDA not only considers the distribution adaptation between domains but also adaptively changes the weight of each class. To evaluate the proposed methods, we conduct extensive experiments on several transfer learning tasks, which demonstrate the effectiveness of our proposed algorithms over several state-of-the-art methods.

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。

↑