含信息保护的跨介质多域保似学习方法及其微生物数据之知识迁移利用研究

In order to overcome the common challenges existing in the.field of knowledge mining and leveraging for cross-media and multi-domain datasets.(for example, microbiological datasets which include text data and/or image data),.novel learning theory and methodology will be investigated in this project. In this.project, the following key issues will be explored: 1) the formal definition of the.concept of cross-media and multi-domain learning problems; 2) The basic algorithms.（i.e., the novel logarithmic-potential similarity-preserving learning machines）and their theoretical analysis and performance evaluation for the cross-media and multi-domain learning; 3) The fast learning algorithms based on the cross-media and.multi-domain similarity-preserving homomorphic dual transform and logarithmic-potential similarity-preserving learning machines for the large-scale cross-media and multi-domain learning problems and their theoretical analysis and.performance evaluation; 4) The distributed learning algorithms for the cross-media.and multi-domain learning under the multi-soure or distributed environment and their.theoretical analysis and performance evaluation; 5) The information protection mechanisms for the cross-media and multi-domain learning; 6) The knowledge leveraging of large-scale microbial data resources by using the proposed cross-media.and multi-domain learning approach. The team's recent researches in intelligent recognition and microbiological information technologies provide the sufficient preparation of this project's feasibility.

本课题拟针对跨介质多域数据的知识迁移利用研究中所涉及的共性挑战问题，探究相应的新的学习理论和方法，并应用于微生物数据资源的知识挖掘与利用研究。归纳起来，本课题紧密结合智能识别，知识挖掘与微生物信息技术，拟探讨如下的几个关键问题：1)跨介质多域学习的问题定义及其表述的规范化；2)跨介质多域保似学习基础算法即对数势保似学习机等及其理论分析与性能评估；3)基于多域保似同态对偶变换机制和对数势保似学习机的大规模数据环境下跨介质多域可扩展快速学习算法及其理论分析与性能评估；4)分布式环境下的跨介质多域学习算法及其理论分析与性能评估；5)跨介质多域学习的信息保护机制研究；6)基于跨介质多域学习的大规模微生物发酵数据资源的知识迁移利用。课题组的前期工作为本课题的可行性提供了充分的准备。

在2013.01-2016.12执行国家自然科学基金（No. 61272210）过程中，按计划开展了相关的研究，重点是含信息保护的跨介质多域保似学习方法及其微生物数据之知识迁移利用研究，并在此基础上在相关方向进行了拓展研究，主要内容研究结果具体包括：针对有监督学习方式，探讨了面向大样本学习问题的跨介质多域保似学习策略及其快速学习方法；针对无监督学习方式，探讨了具备跨介质多域保似学习能力的新型聚类分析方法；针对信息安全和隐私保护问题，探讨了具备信息保护能力的跨介质多域保似学习策略及其学习方法。项目执行过程中在相关领域形成了不少研究成果，其对智能识别理论及其在微生物数据建模等方面的应用具有重要价值。