面向认知的多源数据学习理论与算法

In the age of big data, learning from multi-source data plays an important role in many real applications including social computing, bioinformatics, clinical medicine, robot control, e-commerce, transportation and etc. To date, plenty of multi-source data learning algorithms have been proposed, however, few work focus on the fundamental theoretic laws. Meanwhile, it is hard for the classical machine learning theories (such as computational learning theory, statistical learning theory, probabilistic graphical model) to govern all learning systems, and to further provide a theoretical support for multi-source learning algorithms. Based on the view that learning aims to find knowledge, in this project, we plan to integrate the cognitive mechanism of human category learning, the existing machine learning theories and machine learning models, and focus on key problems as follows: the cognitive mechanism of multi-source data learning, the general representation and axiomatic foundation of machine learning methods. In order to form a cognition-inspired learning framework to govern the typical multi-source learning algorithms and design new learning algorithms, we will propose a category representation theory which can unify the existing cognitive category theories (including classical theory, prototype theory, exemplar theory, knowledge theory, etc.); we will propose a novel machine learning axiomatic system which not only is consistent with human cognitive mechanism but also can govern the existing machine learning theory and algorithms and further supervise to design and evaluate multi-source learning algorithms. Based on these theories and algorithms, we will develop a theoretically supported multi-source health monitoring prototype system for highspeed railway tracks.

多源数据学习在大数据时代具有极其重要意义。目前多源数据学习算法研究远远超前于多源数据学习理论研究，经典的机器学习理论难以应用于多源数据学习，更难以提供多源数据学习算法在高风险应用中的理论保障。从学习的最终目的是知识这一认知切入点出发，本课题基于人类学习的认知机理、机器学习三大经典理论（计算学习理论、统计学习理论和概率图理论）以及典型的机器学习模型，围绕多源数据学习的认知机理、机器学习算法表示的统一化、机器学习公理化等诸科学问题开展研究。提出一个能够统一表示现今概念认知理论（包括经典理论、原型理论、样例理论、知识理论等）的概念表示理论，提出一个可以覆盖典型机器学习算法又与人类认知机理一致的、原创的机器学习公理化体系，不仅可以用来解释现存的机器学习算法与理论，也可以用来指导多源数据学习算法的设计与评估。并基于以上理论和方法，研制出一个理论上有一定保障的高铁轨道多源健康检测原型系统。

多源数据学习在大数据时代具有极其重要意义。目前多源数据学习算法研究远远超前于多源数据学习理论研究，经典的机器学习理论难以应用于多源数据学习，更难以提供多源数据学习算法在高风险应用中的理论保障。从学习的最终目的是知识这一认知切入点出发，本课题基于人类学习的认知机理、机器学习三大经典理论（计算学习理论、统计学习理论和概率图理论）以及典型的机器学习模型，围绕多源数据学习的认知机理、机器学习算法表示的统一化、机器学习公理化等诸科学问题开展研究。提出一个能够统一表示现今概念认知理论（包括经典理论、原型理论、样例理论、知识理论等）的概念表示理论，提出一个可以覆盖典型机器学习算法又与人类认知机理一致的、原创的机器学习公理化体系，不仅可以用来解释现存的机器学习算法与理论，也可以用来指导多源数据学习算法的设计与评估。并基于以上理论和方法，研制出一个理论上有一定保障的高铁轨道多源健康检测原型系统。