| 摘要: |
| 在数据标注任务中，数据标注噪声问题普遍存在，给实际应用带来挑战，因此偏多标记学习作为弱监督学习的重要方向，具有提升标注准确性和降低人工成本的意义。本文提出了一种基于深度森林架构的偏多标记学习模型——偏多标记深度森林（PMLDF），该模型融合了度量敏感的多标记深度森林和偏标记森林的优势，有效缓解候选标签不确定和噪声干扰问题。实验结果表明，PMLDF在多个真实及合成数据集上均展现出优异的泛化能力和鲁棒性，拓展了深度森林架构在弱监督学习领域的应用范畴，为多标签学习及相关应用提供了新的解决思路。 |
| 关键词: 机器学习 偏多标记学习 弱监督学习 深度森林 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Partial Multi-Label Learning method based on deep forest |
|
yuefan, qiufeng
|
|
Shanghai Normal University
|
| Abstract: |
| In the task of data annotation, the problem of data annotation noise is widespread, which brings challenges to the practical application. Therefore, as an important direction of weakly supervised learning, Partial Multi-Label Learning has the significance of improving annotation accuracy and reducing labor costs. The paper proposes a Partial Multi-Label Learning model based on the deep forest architecture, the multi label deep forest (PMLDF). This model combines the advantages of the Measure-Aware Multi-Label Deep Forest and the Partial Label Forest, and effectively alleviates the uncertainty of candidate labels and noise interference. The experimental results show that PMLDF has excellent generalization ability and robustness on multiple real and synthetic datasets, which expands the application scope of deep forest architecture in the field of weak supervised learning, and provides a new solution for multi label learning and related applications. |
| Key words: Machine learning Partial Multi Label Learning Weakly supervised learning Deep forest |
