电子科技大学学报
電子科技大學學報
전자과기대학학보
Journal of University of Electronic Science and Technology of China
2015年
6期
899-904
,共6页
多标签学习%稀疏近邻表示%LASSO稀疏最小化%非负重构
多標籤學習%稀疏近鄰錶示%LASSO稀疏最小化%非負重構
다표첨학습%희소근린표시%LASSO희소최소화%비부중구
LASSO sparse minimization%multi-label learning%non-negative reconstruction%sparse neighbor representation
针对训练数据中的非线性流形结构以及基于稀疏表示的多标签分类中判别信息丢失严重的问题,该文提出一种非负稀疏近邻表示的多标签学习算法。首先找到待测试样本每个标签类上的k-近邻,然后基于LASSO稀疏最小化方法,对待测试样本进行非负稀疏线性重构,得到稀疏的非负重构系数。再根据重构误差计算待测试样本对每个类别的隶属度,最后实现多标签数据分类。实验结果表明所提出的方法比经典的多标签k近邻分类(ML-KNN)和稀疏表示的多标记学习算法(ML-SRC)方法性能更优。
針對訓練數據中的非線性流形結構以及基于稀疏錶示的多標籤分類中判彆信息丟失嚴重的問題,該文提齣一種非負稀疏近鄰錶示的多標籤學習算法。首先找到待測試樣本每箇標籤類上的k-近鄰,然後基于LASSO稀疏最小化方法,對待測試樣本進行非負稀疏線性重構,得到稀疏的非負重構繫數。再根據重構誤差計算待測試樣本對每箇類彆的隸屬度,最後實現多標籤數據分類。實驗結果錶明所提齣的方法比經典的多標籤k近鄰分類(ML-KNN)和稀疏錶示的多標記學習算法(ML-SRC)方法性能更優。
침대훈련수거중적비선성류형결구이급기우희소표시적다표첨분류중판별신식주실엄중적문제,해문제출일충비부희소근린표시적다표첨학습산법。수선조도대측시양본매개표첨류상적k-근린,연후기우LASSO희소최소화방법,대대측시양본진행비부희소선성중구,득도희소적비부중구계수。재근거중구오차계산대측시양본대매개유별적대속도,최후실현다표첨수거분류。실험결과표명소제출적방법비경전적다표첨k근린분류(ML-KNN)화희소표시적다표기학습산법(ML-SRC)방법성능경우。
In order to avoid the influence of the nonlinear manifold structure in training data and preserve more discriminant information in the sparse representation based multi-label learning, a new multi-label learning algorithm based on non-negative sparse neighbor representation is proposed. First of all, thek-nearest neighbors among each class are found for the test sample. Secondly, based on non-negative the least absolute shrinkage and selectionator operator (LASSO)-type sparse minimization, the test sample is non-negative linearly reconstructed by thek-nearest neighbors. Then, the membership of each class for the test sample is calculated by using the reconstruction errors. Finally, the classification is performed by ranking these memberships. A fast iterative algorithm and its corresponding analysis of converging to global minimum are provided. Experimental results of multi-label classification on several public multi-label databases show that the proposed method achieves better performances than classical ML-SRC and ML-KNN.