Abstract: Partial label regression (PLR) complements the current situation that partial label learning (PLL) only discusses on classification tasks. To address the problem that existing PLR algorithms ignored feature differences of instances, a partial label regression algorithm integrating feature attention and residual connection (PLR-FARC) was proposed. Firstly, labels of real datasets were expanded into a set of real-value candidate labels by label enhancement technique. Secondly, the attention mechanism was employed to explore contribution of individual features to labels. Thirdly, the residual connection was introduced to reduce information loss and maintain features integrity during feature transmission. Finally, prediction loss was calculated based on identification method (IDent) and progressive identification method (PIDent), respectively. Experiments were conducted on Abalone, Airfoil, Concrete, Cpu-act, Housing and Power-plant datasets. Compared to IDent and PIDent, the mean absolute error of PLR-FARC is reduced by an average of 2.15%, 38.38%, 8.86%, 4.19%, 15.71% and 15.55%, while the mean squared error is reduced by an average of 9.35%, 71.32%, 23.10%, 20.17%, 27.22% and 9.46%. Experimental results show that the proposed algorithm is feasible and effective.

Key words: partial label learning, partial label regression, candidate label, attention mechanism, residual connection

摘要： 偏标签回归(PLR)补全了偏标签学习(PLL)仅讨论分类任务的现状。针对现有PLR算法忽略实例特征特性差异问题，提出了一种集成特征注意力和残差连接的偏标签回归算法(PLR-FARC)。首先，通过标签增强技术将真实数据集的标签扩充为一组实值候选标签；其次，借助注意力机制自动生成每个特征对标签的贡献度；再次，引入残差连接以减少特征在传递过程中的信息丢失，维持特征的完整性；最后，分别基于IDent和PIDent计算预测损失。在Abalone、Airfoil、Concrete、Cpu-act、Housing和Power-plant数据集上进行实验，相较于IDent和PIDent，PLR-FARC的平均绝对误差分别平均降低了2.15%、38.38%、8.86%、4.19%、15.71%和15.55%，均方误差分别平均降低了9.35%、71.32%、23.10%、20.17%、27.22%和9.46%。实验结果表明，所提算法是可行且有效的。

关键词: 偏标签学习, 偏标签回归, 候选标签, 注意力机制, 残差连接