Disjoint Label Space Transfer Learning with Common Factorised Space
Disjoint Label Space Transfer Learning with Common Factorised Space
复制标题
DOI：
10.1609/aaai.v33i01.33013288
复制
发表时间：
2018-12
期刊：
影响因子：
--
通讯作者：
Xiaobin Chang;Yongxin Yang;T. Xiang;Timothy M. Hospedales
中科院分区：
文献类型：
--
作者：
Xiaobin Chang;Yongxin Yang;T. Xiang;Timothy M. Hospedales
在本文中，提出了一种统一的迁移学习方法，该方法使用单个模型解决多个源域和目标域标签空间和注释假设。它在处理源标签空间和目标标签空间不相交的具有挑战性的情况下特别有效，并且在无监督和半监督设置中都优于替代方案。关键要素是称为公共因式分解空间的通用表示。它在源域和目标域之间共享，并使用无监督分解损失和基于图的损失进行训练。通过广泛的实验，我们证明了我们的方法的灵活性、相关性和有效性，无论是在标签空间不相交的具有挑战性的情况下，还是在更传统的情况下，例如无监督域适应，其中源域和目标域共享相同的标签集。
In this paper, a unified approach is presented to transfer learning that addresses several source and target domain labelspace and annotation assumptions with a single model. It is particularly effective in handling a challenging case, where source and target label-spaces are disjoint, and outperforms alternatives in both unsupervised and semi-supervised settings. The key ingredient is a common representation termed Common Factorised Space. It is shared between source and target domains, and trained with an unsupervised factorisation loss and a graph-based loss. With a wide range of experiments, we demonstrate the flexibility, relevance and efficacy of our method, both in the challenging cases with disjoint label spaces, and in the more conventional cases such as unsupervised domain adaptation, where the source and target domains share the same label-sets.