.png){kind=icon}
Return
Meta-learning-based multi-view multi-label feature selection with streaming labels
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.116084.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-view multi-label feature selection plays a pivotal role in handling high-dimensional heterogeneous data. However, most existing approaches rely on the assumption of a static label space, requiring all class labels to be available prior to training. This limitation renders them ill-equipped for real-world scenarios where labels arrive continuously in a streaming fashion. While Streaming Label Feature Selection addresses the dynamic expansion of label spaces, existing algorithms often overlook the complementary information in multi-view data. To address these challenges, we propose a Meta-learning-based multi-view multi-label Feature Selection with Streaming Labels (MFSSL) method. We formulate streaming label learning as a continuous meta-learning process, employing a gradient-based meta-optimization strategy to achieve fast adaptation to incoming label batches. Specifically, we design a historical stability constraint mechanism that regularizes the current feature mask update using historical memory, thereby effectively preserving valuable historical patterns while maintaining plasticity for new labels. Furthermore, to handle class imbalance and improve feature quality, we introduce a discriminative margin constraint to directly optimize the discrimination capability of the selected features. Extensive experiments on multiple benchmark datasets demonstrate that MFSSL outperforms state-of-the-art methods in terms of classification accuracy and computational efficiency in streaming label environments.
Keywords:
Meta-learning
Multi-view feature selection
Streaming labels
Label space expansion
Feature selection
Journal
K
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
View more organizations
Cited Papers
No cited papers available