摘要
为充分利用过期训练数据和数据结构相关性进行新领域的学习,提出一种基于Markov逻辑网的迁移学习方法。该方法对源域与目标域的谓词进行自动映射后,通过自我诊断、结构更新和新公式挖掘3个步骤对映射结构进行优化,使之更适用于目标域数据。实验结果证明,与传统的机器学习方法相比,该方法使概率推理所获结果的准确率更高,所需的学习时间与训练数据更少。
In order to take advantages of expired training data and correlation data structures to achieve the aim of learning new areas, this paper proposes a transfer learning approach based on Markov Logic Network(MLN). It autonomously maps the predicates on the source and target, and optimizes the mapping structure by self-diagnosis, structure update and new formula mining. Experimental results show that compared with traditional method, the probabilistic reasoning to the target domain MLN structure to gain higher accuracy of the results with less learning time and training data.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第24期158-160,共3页
Computer Engineering
基金
中央高校研究生科技创新基金资助项目(CDJXS11180013)
关键词
迁移学习
MARKOV逻辑网
自动映射
机器学习
一阶逻辑
transfer learning
Markov Logic Network(MLN)
autonomous mapping
machine learning
first-order logic