贺建军,王欣,顾宏,王哲龙.基于Logistic回归模型和凝聚函数的多示例学习算法[J].,2010,(5):788-793 |
基于Logistic回归模型和凝聚函数的多示例学习算法 |
An algorithm for multi-instance learning based on Logistic regression model and aggregate function |
|
DOI:10.7511/dllgxb201005029 |
中文关键词: 多示例学习 Logistic回归模型 凝聚函数 文本分类 |
英文关键词: multi-instance learning Logistic regression model aggregate function text categorization |
基金项目:国家地震行业科研专项基金资助项目(200808075). |
|
摘要点击次数: 1228 |
全文下载次数: 1029 |
中文摘要: |
鉴于很多实际问题都可以转化到多示例框架下求解,多示例学习越来越受到机器学习领域内学者们的关注.提出了一个基于Logistic回归模型的多示例学习算法.首先定义了一个新的似然函数来表示每个包的标签与其示例的隐含标签之间的关系,然后利用凝聚函数把该似然函数转化为一个光滑的凹函数,从而使问题可以用常用的无约束优化方法快速求解.在一些标准数据集和一个文本分类问题上的实验结果表明,所提算法要优于其他常用多示例学习算法. |
英文摘要: |
Since many real-world problems could be transformed and solved in multi-instance learning (MIL) framework, the researchers in the field of machine learning have paid more and more attention to MIL. Based on the Logistic regression model, a new MIL algorithm is developed. Firstly, in this algorithm, a new likelihood function is defined to build the relationship between the bag′s class label and its instances′ hidden class labels. Then, the aggregate function is used to transform the likelihood function to a smooth concave function, thereby the problem can be solved by a general unconstrained optimization method. Experimental results of both the benchmark data sets and a text categorization problem show that the proposed algorithm can achieve superior performance to the published MIL algorithms. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |
|
|
|