可加性特征涌现的外延模糊逻辑识别系统及其基于锚点学习机的快速学习与应用

Data is sometimes collected with the addition of emerging features in practical applications, which heavily challenges the classical data driven intelligent recognition system modeling, from the perspective of dynamic epitaxial features, scene migration and learning of large scale data with epitaxial features. In view of the above challenges, the research group will (1)propose the new concept of epitaxial fuzzy logic systems for recognition and modeling tasks and then investigate how to construct the proposed epitaxial fuzzy logic system and analyse its theoretical properties; (2) investigate how to construct the proposed transfer epitaxial fuzzy logic system for the transfer learning scenes; (3) investigate how to design the robust versions of the above proposed models and discuss their theoretical properties; (4) propose the anchoring vector machine for fast modeling on large (noisy) datasets, and then in terms of the proposed anchoring vector machine, develop the fast training algorithms of the above proposed models for large (noisy) datasets. Finally, the proposed methods and techniques will be applied to microorganism fermentation process quality evaluation along with the evolving data with the epitaxial features in order to validate the feasibility and effectiveness of the proposed models. The existing research advances achieved by the research group provide solid foundation for this research topic.

本课题以具有可加性特征涌现的数据建模为需求面向，针对其给经典的数据驱动的智能识别系统构建所带来的一系列新的诸如特征动态外延、场景迁移以及具有特征外延特点的大规模数据学习等挑战，首先提出外延模糊逻辑识别系统，并研究其构建方法以及理论性质；针对因可加性特征涌现可能带来的场景迁移，研究基于迁移学习的外延模糊逻辑识别系统的构建方法及其理论性质；研究面向噪音数据的鲁棒外延模糊逻辑识别系统的构建方法及其理论性质；提出适合大规模(含噪)数据的锚点学习机，以及基于锚点学习机的(鲁棒)外延模糊逻辑识别系统高效构建与学习技术。最终课题组拟把提出的外延模糊逻辑识别系统构建技术应用于微生物发酵过程质量评估等方面，以验证所提出的外延模糊逻辑识别系统的可用性和有效性。课题组的前期研究成果为本课题的开展提供了充分的准备。

在国家自然科学基金面上项目（可加性特征涌现的外延模糊逻辑识别系统及其基于锚点学习机的快速学习与应用，批准号： 61572236）的资助下，按计划开展了相关的研究，以具有可加性特征涌现的数据建模需求为面向，针对因可加性特征涌现可能带来的场景迁移、噪音数据、大规模数据等挑战，研究了基于迁移学习的外延模糊逻辑识别系统、鲁棒外延模糊逻辑识别系统以及快速学习模糊逻辑识别系统高效构建。相关成果发表和录用学术论文20篇，其中IEEE Trans. 系列长文13篇，获得各级奖励1项，申请专利3项，获得软件著作权1项。依托该项目，2人次获得各级人才项目和人才计划资助。