针对基于传感器的行为识别任务中识别场景单一且固定的问题,提出一种多场景下基于传感器的行为识别迁移模型,由基于传感器的动态感知算法(dynamic perception algorithm,DPA)和自适应场景的行为识别迁移方法(adaptive scene human recog...针对基于传感器的行为识别任务中识别场景单一且固定的问题,提出一种多场景下基于传感器的行为识别迁移模型,由基于传感器的动态感知算法(dynamic perception algorithm,DPA)和自适应场景的行为识别迁移方法(adaptive scene human recognition,AHR)两部分组成,解决在固定场景下对传感器的依赖性以及在场景转换时识别模型失效的问题。DPA提出两阶段迁移模式,将行为识别阶段和模型迁移阶段同步推进,保证模型在传感器异动发生后仍能持续拥有识别能力。进一步提出AHR场景迁移方法,实现模型在多场景下的行为识别能力。实验验证该模型具有更优的适应性和可扩展性。展开更多
Using Java, Java enabled Web and object-oriented programming technologies, a framework is designed to or ganize multicornputer system on lntranet quickly to complete Monte Carlo simulation parallelizing, The high-perf...Using Java, Java enabled Web and object-oriented programming technologies, a framework is designed to or ganize multicornputer system on lntranet quickly to complete Monte Carlo simulation parallelizing, The high-performance computing enviromnent is embedded in Web server so it can be accessed more easily. Adaptive parallelism and eager scheduling algorithm are used to realize load balancing, parallel processing and system fault-tolerance. Independent sequence pseudo-randorn number generator schemes to keep the parallel simulation availability. Three kinds of stock option pricing models as instances, ideal speedup and pricing results obtained on test bed. Now, as a Web service, a high-performance financial derivative security-pricing platform is set up for training and studying. The framework can also be used to develop other SPMD (single procedure multiple data) application. Robustness is still a major problem for further research.展开更多
文摘针对基于传感器的行为识别任务中识别场景单一且固定的问题,提出一种多场景下基于传感器的行为识别迁移模型,由基于传感器的动态感知算法(dynamic perception algorithm,DPA)和自适应场景的行为识别迁移方法(adaptive scene human recognition,AHR)两部分组成,解决在固定场景下对传感器的依赖性以及在场景转换时识别模型失效的问题。DPA提出两阶段迁移模式,将行为识别阶段和模型迁移阶段同步推进,保证模型在传感器异动发生后仍能持续拥有识别能力。进一步提出AHR场景迁移方法,实现模型在多场景下的行为识别能力。实验验证该模型具有更优的适应性和可扩展性。
基金Supported by the National 863 High TechnologyDevelopment of China (2001AA111081)
文摘Using Java, Java enabled Web and object-oriented programming technologies, a framework is designed to or ganize multicornputer system on lntranet quickly to complete Monte Carlo simulation parallelizing, The high-performance computing enviromnent is embedded in Web server so it can be accessed more easily. Adaptive parallelism and eager scheduling algorithm are used to realize load balancing, parallel processing and system fault-tolerance. Independent sequence pseudo-randorn number generator schemes to keep the parallel simulation availability. Three kinds of stock option pricing models as instances, ideal speedup and pricing results obtained on test bed. Now, as a Web service, a high-performance financial derivative security-pricing platform is set up for training and studying. The framework can also be used to develop other SPMD (single procedure multiple data) application. Robustness is still a major problem for further research.