多标记学习自编码网络无监督维数约简

多标记学习是针对一个实例同时与一组标签相关联而提出的一种机器学习框架，是该领域研究热点之一，降维是多标记学习一个重要且具有挑战性的工作。针对有监督的多标记维数约简方法，提出一种无监督自编码网络的多标记降维方法。首先，通过构建自编码神经网络，对输入数据进行编码和解码输出；然后，引入稀疏约束计算总体成本，使用梯度下降法进行迭代求解；最后，通过深度学习训练获得自编码网络学习模型，提取数据特征实现维数约简。实验中使用多标记算法ML-kNN做分类器，在6个公开数据集上与其他4种方法对比。实验结果表明，该方法能够在不使用标记的情况下有效提取特征，降低多标记数据维度，稳定提高多标记学习性能。

Unsupervised dimensionality reduction of multi-label learning via autoencoder networks

Multi-label learning is a machine learning framework that simultaneously deals with data associated with a group of labels. It is one of the hot spots in the field of machine learning. However, dimensionality reduction is a significant and challenging task in multi-label learning. In this paper, we propose a unsupervised dimensionality reduction method for supervised multi-label dimensiona reduction methods via autoencoder networks. Firstly, we build the autoencoder neural network to encode the input data and then decode them for output. Then we introduce the sparse constraint to calculate the overall cost, and further use the gradient descent method to iterate them. Finally, we obtain the autoencoder network learning model by deep learning training, and then extract data features to reduce dimensionality. In the experiments, we use a multi-label algorithm (ML-kNN) as the classifier, and compare them with four other methods on six publicly available datasets. Experimental results show that the proposed method can effectively extract features without using label learning; thus, it reduces multi-label data dimensionality and steadily improves the performance of multi-label learning.