[论文解读] Domain Separation Networks

Konstantinos Bousmalis, George Trigeorgis+3 more|arXiv (Cornell University)|Aug 22, 2016

Domain Adaptation and Few-Shot Learning参考文献 31被引用 588

一句话总结

DSN 显式建模私有（域特定）和共享表示用于无监督域自适应，结合重构和正交损失以及相似性引导，优于现有方法。

ABSTRACT

The cost of large scale data collection and annotation often makes the application of machine learning algorithms to new tasks or datasets prohibitively expensive. One approach circumventing this cost is training models on synthetic data where annotations are provided automatically. Despite their appeal, such models often fail to generalize from synthetic to real images, necessitating domain adaptation algorithms to manipulate these models before they can be successfully applied. Existing approaches focus either on mapping representations from one domain to the other, or on learning to extract features that are invariant to the domain from which they were extracted. However, by focusing only on creating a mapping or shared representation between the two domains, they ignore the individual characteristics of each domain. We suggest that explicitly modeling what is unique to each domain can improve a model's ability to extract domain-invariant features. Inspired by work on private-shared component analysis, we explicitly learn to extract image representations that are partitioned into two subspaces: one component which is private to each domain and one which is shared across domains. Our model is trained not only to perform the task we care about in the source domain, but also to use the partitioned representation to reconstruct the images from both domains. Our novel architecture results in a model that outperforms the state-of-the-art on a range of unsupervised domain adaptation scenarios and additionally produces visualizations of the private and shared representations enabling interpretation of the domain adaptation process.

研究动机与目标

• 通过利用合成数据和领域自适应来降低标注成本的动机。
• 提出一个两子空间表示：私有（域特定）和共享（域不变）。
• 在源域上以任务损失进行训练，并在两个域上添加重构损失和正交损失以实现表示分离。
• 在自适应过程中实现对私有和共享组件的可视化/可解释性成为可能。

提出的方法

• 为每个域引入共享编码器 Ec 和私有编码器 Ep。
• 使用一个共享解码器 D，从 Ec 和 Ep 表示重建输入。
• 施加差异损失以鼓励私有和共享组件之间的正交性（L_difference）。
• 施加相似度损失以在跨域对齐共享表示（L_similarity），使用 DANN（带 GRL）或 MMD。
• 在源标签的任务损失和两个域的重构损失下进行训练；在预热阶段后应用领域自适应损失。
• 跨多个域自适应场景共享架构和训练流程。

实验结果

研究问题

• RQ1 如何通过显式建模域私有和域共享表示来提升无监督域自适应？
• RQ2 私有-共享分区和重构损失是否比先前的映射或共享表示方法在跨域泛化上表现更好？
• RQ3 哪种相似性目标（DANN 与 MMD）最有助于在 DSN 内实现跨域对齐？
• RQ4 DSN 是否能够在自适应过程中提供私有与共享表示的可解释可视化？

主要发现

模型	MNIST-M	SVHN	MNIST	GTSRB
Source-only	56.6 (52.2)	86.7 (86.7)	59.2 (54.9)	85.1 (79.0)
CORAL [27]	57.7	85.2	63.1	86.9
MMD [30,18]	76.9	88.0	71.1	91.1
DANN [8]	77.4 (76.6)	90.3 (91.0)	70.7 (73.8)	92.9 (88.6)
DSN w/ MMD (ours)	80.5	88.5	72.2	92.6
DSN w/ DANN (ours)	83.2	91.2	82.7	93.1
Target-only	98.7	92.4	99.5	99.8

• 使用 DANN 相似性损失的 DSN 在多项无监督域自适应任务中优于 CORAL、MMD 以及标准 DANN 基线。
• 在消融实验中，去除差异（正交性）损失会在所有任务上降低性能。
• 用标准 L2 MSE 替换尺度不变重构会降低性能，验证了提出的重构损失。
• DSN 在 MNIST→MNIST-M、Synth Digits→SVHN、SVHN→MNIST，以及 Synth Signs→GTSRB 任务上获得高准确率。
• DSN 在 Synth Objects→LINEMOD 场景也获得具有竞争力的姿态估计结果。