The Research On Transdctive Transfer Learning With The Logistic Regression Model

Traditional data mining techniques make a basic assumption that the training and test data should be under the same distributions. However, in many cases, this identical-distribution assumption does not hold. When the distribution changes, most traditional data mining methods need to be rebuilt from scratch using newly collected training data. In many real world applications, it is expensive or impossible to re-collect the needed training data and rebuild the models. In this situation, transfer learning becomes important to take the role of leveraging these existing data knowledge.Transfer learning aims to extract the knowledge from one or more source tasks and applies the knowledge to a target task. The study of Transfer learning is motivated by the fact that people can intelligently apply knowledge learned previously to solve new problems faster or with better solutions. Thus, it does not make the identical-distribution assumption as tractional data mining algorithms.The work if this dissertation aims at the transfer learning problems, and the main contributions are as follows:(1) The whole field of transfer learning are reviewed, and the existing algorithms for the transfer learning are summarized and analyzed.(2) In order to address the target domian which have no labeled data, an novel method (TTLR) which is based on the logistic regression model for transductive transfer learning is proposed. The learned model is more adaptive for the target domain when the training data and the testing data does not under the same distributions. The results of experiments indicate that TTLR outperforms the triditional supervised learning method.(3) In order to deal with multiple different distributions simultaneously, a method (MTTLR) for multi-domain for transductive transfer learning is proposed. The results of experiments show that MTTLR has an improved performance in accuracy in comparison with the triditional supervised learning method and some existint algorithms for the transductive transfer learning.