数据流半监督分类中的半监督迁移学习研究

In the fields of information security, automotive manufacturing, and social culture et al, there are many urgent demands for classifying data streams accurately to ensure quality and safety. In these sceneries, however, all instances collected cannot be labeled immediately for the reasons of cost, time, manpower, user’s habit, and sometimes the true label really cannot be found easily. These difficulties bring many challenges to semi-supervised classification of the data streams. In order to overcome these challenges, this project aims to explore a semi-supervised transfer learning algorithm which can transfer knowledge from multiple source domains to a target domain for semi-supervised classification of data streams, on the conditions of source domain and target domain are all semi-supervised. .The research plan includes that: semi-supervised clustering based classification, semi-supervised transfer learning, mutual interaction between the knowledge extracted from source domain and target domain, detection concept drifts based on the relation between source domain and target domain, and apply all the proposed algorithm to handle network intrusion, industrial image detection, spam filtering, and products evaluation. If this proposal can be supported, it will inspire designing new machine learning models like adaptive learning, lifelong machine learning guided by the mechanism of cognitive learning, and make these works much more significantly.

在信息安全、自动化生产和社会文化等领域，为确保质量和安全，迫切需要对数据流准确分类。但由于成本、时间、人手、使用习惯等原因，再加上有时的确很难正确地标注样本，从而无法给所有样本都实时标注其真实类别。在这类场景下，数据流的半监督分类面临重重挑战。本项目重点研究一种可用于数据流半监督分类的半监督迁移学习算法。该算法能在源领域和目标领域都是半监督的条件下，实现从多个源领域到目标领域的知识迁移。研究内容主要包括——基于半监督聚类的分类器、半监督迁移学习、源领域和目标领域知识之间的相互作用、基于源领域与目标领域相关性的概念漂移检测，并将以上算法用于解决网络入侵检测、工业图像检测、垃圾邮件过滤和产品评价等实际问题。本项目对于根据认知学习机制来研究自适应学习和终身机器学习等新的机器学习模型具有重要的科学意义。

本项目针对信息安全、自动化生产和社会文化等领域数据流半监督分类问题，重点研究一种可用于数据流半监督分类的半监督迁移学习算法。该算法能在源领域和目标领域都是半监督的条件下，实现从多个源领域到目标领域的知识迁移。研究内容主要有——基于半监督聚类的分类器、半监督迁移学习、源领域和目标领域知识之间的相互作用、基于源领域与目标领域相关性的概念漂移检测，并将以上算法用于解决网络入侵检测等实际问题。本项目对于网络空间安全、智能制造领域意义重大，对于根据认知学习机制来研究自适应学习和终身机器学习等新的机器学习模型也具有重要的科学意义。