Feature Decoupling
Feature decoupling in machine learning aims to separate entangled aspects of data representations, improving model performance and interpretability. Current research focuses on decoupling features within various contexts, including modality-specific and modality-shared information in multi-modal learning, geometric and semantic features in 3D scene understanding, and content and style features in image generation. This approach leads to more robust and accurate models across diverse applications, such as medical image analysis, autonomous driving, and natural language processing, by addressing limitations of traditional methods that treat data holistically.
Papers
Multi-Dimensional Refinement Graph Convolutional Network with Robust Decouple Loss for Fine-Grained Skeleton-Based Action Recognition
Sheng-Lan Liu, Yu-Ning Ding, Jin-Rong Zhang, Kai-Yuan Liu, Si-Fan Zhang, Fei-Long Wang, Gao Huang
Hierarchical Dense Correlation Distillation for Few-Shot Segmentation-Extended Abstract
Bohao Peng, Zhuotao Tian, Xiaoyang Wu, Chengyao Wang, Shu Liu, Jingyong Su, Jiaya Jia
Taking A Closer Look at Visual Relation: Unbiased Video Scene Graph Generation with Decoupled Label Learning
Wenqing Wang, Yawei Luo, Zhiqing Chen, Tao Jiang, Lei Chen, Yi Yang, Jun Xiao
Retrieval-Augmented Classification with Decoupled Representation
Xinnian Liang, Shuangzhi Wu, Hui Huang, Jiaqi Bai, Chao Bian, Zhoujun Li