摘要
传统的Logistic Regression能够解决单一模态数据的二分类问题,但在处理多源异构数据时不能很好地利用不同模态间的语义相关性,从而降低了分类性能。为了对双模态数据进行建模,提出同时包含模态内语义信息和模态间语义相关性的双模态Logistic Regression模型。设计一个包含模态内损耗与模态间损耗的目标函数,利用梯度下降法优化目标函数,在每次迭代过程中该模型能够根据一定策略交替地更新不同模态的参数。实验结果表明,双模态Logistic Regression能够获得较好的分类性能和跨模态检索效果。
Traditional Logistic Regression can solve the binary classification problem of single modal data,but it cannot make good use of semantic correlation among different modalities when dealing with heterogeneous data,which reduces the classification performance.To model bimodal data,a Bimodal Logistic Regression model is proposed,which captures both intra-modal semantic information and inter-modal semantic correlation.An objective function,which simultaneously considers the intra-modal and inter-modal losses,was designed and then optimized it by gradient descent.In each iteration process,the model can update the parameters of different modes alterhately according to a certain strategy.The experimental results show that the Bimodal Logistic Regression can obtain better classification performance and cross-modal retrieval results.
作者
吴蕊
孔前进
王世勋
孙东山
翟怡星
Wu Rui;Kong Qianjin;Wang Shixun;Sun Dongshan;Zhai Yixing(College of Computer and Information Engineering,Henan Normal University,Xinxiang 453007,Henan,China;Engineering Lab of Intelligence Business and Internet of Things,Henan Normal University,Xinxiang 453007,Henan,China)
出处
《计算机应用与软件》
北大核心
2020年第12期244-248,333,共6页
Computer Applications and Software
基金
国家自然科学基金项目(11702087)
河南省自然科学基金项目(162300410177)
河南省高等学校重点科研项目(17A520040)
河南师范大学博士科研启动基金项目(qd15134)。
