Multi kernel method improves online multi label classification accuracy
Optimal Multi-Kernel Combination for Online Multi-Label Classification
Machine LearningComputer Vision and Pattern Recognition
Summary
Multi-label classification involves assigning multiple labels to data items, which becomes difficult when data arrives one piece at a time, like in a stream. Existing methods usually pick just one way (kernel) to process the data, which might not be the best choice since they can't see all the data at once. The authors designed a new method that combines multiple kernels efficiently, learning how to weigh each kernel’s contribution as new data arrives. This approach leads to better and faster learning, and it eventually matches the best possible fixed combination of kernels chosen after seeing all the data.
What this means in practice
- •For real-time data engineers: Improve multi-label classification accuracy for streaming data applications by combining multiple kernel models adaptively as data arrives.
- •For bioinformatics teams: Enhance online classification of complex biological data with multi-kernel learning to capture diverse data patterns in real-time experiments.
Authors
Tingting Zhai, Wei Liu, Renjian Yu
Abstract
Multi-label classification is particularly challenging in streaming scenarios, where data arrives sequentially. Online kernel methods offer a promising solution. However, prevailing methods rely on a single preselected kernel, making them prone to suboptimal choices since the global data distribution is unknown in advance in online settings. Although the recent method, Online Kernel Selection for Online Multi-label Classification (OKS-OMC), enables data-driven kernel selection, it fails to fully harness the complementary information from multiple kernels and encounters the curse of kernelization. To address these limitations, this paper introduces a novel online multi-kernel combination method for large-scale multi-label classification. First, an efficient and robust online single-kernel multi-label method is designed. Second, learning an optimal convex combination of multiple multi-label kernel classifiers is formalized as a structural risk minimization problem. An improved alternating optimization procedure is developed, in which the combination coefficients are updated by directly minimizing the objective function, while each kernel classifier is updated by minimizing a surrogate upper bound. This procedure decouples the learning of kernel classifiers from combination coefficients, allowing independent classifier updates via our single-kernel method and accelerating convergence to high-quality solutions. Our multi-kernel method is proven to asymptotically converge to the optimal fixed convex combination of the best fixed single-kernel models chosen in hindsight. Ablation studies confirm that our method's components outperform their counterparts in OKS-OMC. Comprehensive evaluations across 11 datasets and seven metrics demonstrate our method's overall superiority over existing online multi-label methods.