您的位置:首页  > 论文页面

基于正则化稀疏的软谓词发现问题及研究

发表时间:2017-05-31  浏览量:1357  下载量:418
全部作者: 潘敬敏,高志强
作者单位: 东南大学-蒙纳士大学苏州联合研究院;东南大学信息科学与工程学院
摘 要: 软谓词发现(predicate invention,PI)方法并不是明确地创建新谓词,而是通过使用正则化稀疏方式将它们的参数一起正则化,从而隐式地组合紧密相关的规则。提出一种基于弹性网的软PI方法的改进办法,能够有效避免错误级联的发生,且实验表明可大大提高PI方法的性能,并且在大规模学习任务中的表现尤为突出。
关 键 词: 人工智能;谓词发现;正则化稀疏;弹性网;关系学习
Title: Research on soft predicate invention based on regularized sparsity
Author: PAN Jingmin, GAO Zhiqiang
Organization: Southeast University-Monash University Joint Graduate School (Suzhou); School of Information Science and Engineering, Southeast University
Abstract: Instead of explicitly creating new predicates, the soft predicate invention (PI) implicitly groups the closely-related rules by using regularized sparsity to regularize their parameters together. This paper presents an improved method of soft PI based on elastic net, which can effectively avoid the occurrence of error cascades. Results show that it can effectively improve the performance of PI, especially in large-scale learning tasks.
Key words: artificial intelligence; predicate invention; regularized sparsity; elastic net; relational learning
发表期数: 2017年5月第10期
引用格式: 潘敬敏,高志强. 基于正则化稀疏的软谓词发现问题及研究[J]. 中国科技论文在线精品论文,2017,10(10):1106-1116.
 
0 评论数 0
暂无评论
友情链接