计算机科学与探索 ›› 2020, Vol. 14 ›› Issue (7): 1183-1193.DOI: 10.3778/j.issn.1673-9418.1905085

• 人工智能 • 上一篇    下一篇

基于联合信息保持的异构领域自适应

许鹏,邓赵红,王骏,王士同   

  1. 江南大学 人工智能和计算机学院,江苏 无锡 214122
  • 出版日期:2020-07-01 发布日期:2020-08-12

Joint Information Preservation for Heterogeneous Domain Adaptation

XU Peng, DENG Zhaohong, WANG Jun, WANG Shitong   

  1. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2020-07-01 Published:2020-08-12

摘要:

异构领域自适应(HDA)的主要目的是借助源域的知识,辅助处于不同特征空间中目标域的数据进行建模。异构领域自适应一个核心的问题是如何在领域适配过程中有效保持原始数据的信息,减少因为特征变换导致的信息损失,提出了一个联合信息保持算法(JIP)来解决上述问题。所提算法通过两方面来保持原始数据的信息:一方面,在异构领域自适应场景中,两个领域之间通常存在大量配对样本,所提算法通过最大化配对样本之间的相关性来保持这种配对信息。另一方面,所提算法优化了结构信息保持策略,同时保持了原始数据的局部结构信息和全局结构信息。最终,将联合信息保持和分布匹配整合在一起,从而实现异构领域自适应。实验结果表明,所提算法较之于先进的异构领域自适应算法有明显优势。

关键词: 异构领域自适应(HDA), 信息损失, 配对信息

Abstract:

Heterogeneous domain adaptation (HDA) aims to assist the modeling tasks of the target domain lied in different spaces with knowledge of the source domain. A core issue of HDA is how to preserve the information of the original data during domain adaptation and eliminate the information loss caused by feature transformation. This paper proposes a joint information preservation (JIP) method to deal with the problem. The method preserves the information of the original data from two aspects. On the one hand, large number of paired samples often exist between the two domains of the HDA and the proposed method preserves the paired information by maximizing the correlation of the paired samples. On the other hand, the proposed method improves the strategy of preserving the structural information of the original data, where the local and global structural information are preserved simultaneously. Finally, the JIP is integrated with distribution matching to achieve HDA tasks. Experimental results show the superiority of the proposed method over the state-of-the-art HDA algorithms.

Key words: heterogeneous domain adaptation (HDA), information loss, paired information