【摘要】 标记分布学习是一种新型的学习范式,该文提出了一种适用于标记分布问题形式的Adaboost集成算法,能够有效地提升各种单独算法的预测精度。该文设计了一种新的用于反映排序损失的评价指标SortLoss。该文将Adaboost应用在标记分布学习问题上。实验结果表明,该文设计的Adaboost-LDL集成框架在13个真实数据集上能够显著提升标记分布学习算法的预测精度,该文的方法将排序损失指标SortLoss平均降低至原先的41.2%,KL散度指标平均降低至原先的68.5%。更多还原

【Abstract】 Label distribution learning is a new learning paradigm. This paper proposes an Adaboost ensemble algorithm suitable for the form of label distribution learning problem,which can effectively improve the prediction accuracy of various individual algorithms. This paper designs a new evaluation index SortLoss to reflect the sorting loss. This paper applys Adaboost to label distribution learning problems. The experimental results show that the Adaboost-LDL integration algorithm proposed here can significantly improve various existing LDL algorithms on 13 real data sets. Compared with the pre-integration algorithm,the sorting loss can be reduced to an average of 41.2% and the Kullback-Leibler Divergence indicator is reduced to an average of 68.5%.更多还原