Infrastructures, methods, and tools for mining and analyzing personal behavior for innovative recommender and learning systems using hardware accelerators

我们的研究计划的长期目标是开发，建立和传播先进的基础设施，方法和工具，以支持行为挖掘和创新应用分析，包括（i）网络推荐系统;（ii）计算机辅助体育教学系统;和（iii）远程物理治疗和康复系统。特别是，我们将（1）研究捕获，挖掘，跟踪，建模，表示，分类和存档个人行为的理论，方法和技术;（2）使用嵌入式设备在实际应用中实现所产生的方法和技术;（3）使用可编程和可配置的硬件（如FPGA，netFPGA和BEE 3/4）优化这些应用程序和设备。计算机辅助体育教学是人类教练方法的一种有吸引力的替代方案，可以节省大量的空间，时间和成本。我们正在设计一个新的网球电子学习系统，使用游戏控制器作为虚拟球拍，为学习者提供技能改进建议。我们的假设是，从这个电子学习系统的结果可以很容易地应用到康复，特别是远程物理治疗。然后，患者只需在家中遵循规定的运动和个性化建议，而无需访问诊所。因此，该方法包括（a）开发行为分析的理论和方法，（B）开发有效的算法和技术来操纵行为知识，（c）使用硬件加速器优化实现，（d）使用实际应用评估这些技术，并迭代这四个阶段。这项研究是及时的，也是对HQP的一个很好的培训。调查个人行为知识不仅是我们的研究计划所针对的应用程序的基础，也是推荐和电子学习系统的基础。我们的目标应用具有很大的潜力，可以被我们的工业合作伙伴转化为产品。

The long-term objective of our research program is to develop, establish, and disseminate advanced infrastructures, methods, and tools to support behavior mining and analysis for innovative applications including (i) web recommender systems; (ii) computer aided sports instruction systems; and (iii) remote physical therapy and rehabilitation systems. In particular, we will (1) investigate theories, methods, and techniques for capturing, mining, tracking, modeling, representing, classifying, and archiving personal behavior; (2) implement the resulting methods and techniques in practical applications using embedded devices; and (3) optimize these applications and devices using programmable and configurable hardware such as FPGAs, netFPGAs, and BEE3/4. Computer aided sports instruction is an attractive alternative to the human coaching approach, with considerable savings in space, time, and cost. We are designing a novel tennis e-learning system using a game controller as a virtual racquet to provide learners skill improvement recommendations. Our hypothesis is that the results from this e-learning system can be readily applied to rehabilitation and in particular remote physical therapy. A patient then simply follows the prescribed motions and personalized recommendation at home without the need to visit a clinic. Thus, the methodology involves (a) developing theories of and methods for behavior analysis, (b) developing efficient algorithms and techniques to manipulate behavior knowledge, (c) optimizing the implementation using hardware accelerators, (d) evaluating these techniques using practical applications, and iterating over these four phases. This research is timely and a great training ground for HQP. Investigating personal behavior knowledge is not only a foundation for the applications targeted by our research program but also a foundation for recommender and e-learning systems in general. Our target applications have great potential to be turned into products by our industrial partners.