Towards a General Automatic Reasoning Framework for Networked Systems

摘要智能优点：本研究的目的是为网络系统中的自动推理开发一个通用的、灵活的框架，该框架可以广泛应用，并在准确性和效率之间提供所需的权衡，同时允许简单和分布式的实现。鉴于精确干扰通常难以处理，而广泛采用的平均场方法在推理质量上受到限制，本研究的方法是探索结构化变分方法来弥补差距，并允许在各种潜在冲突的目标之间进行灵活的权衡，这些目标可以根据应用程序的需求和可用资源进行调整。在这个变分推理框架中，将考虑各种模型、依赖结构和推理算法，以及高效的分布式实现。一个相关的研究方向是设计能够理想地匹配推理任务需求的自动聚类方案，同时又易于以分布式方式实现。将探索聚类层和推理层之间的交互，以获得更好的性能和自适应性。更广泛的影响：该项目为大规模网络系统（如传感器网络和网络物理系统）中的自动推理构建一般和系统框架提供了基本概念、设计、分析工具和软件。预期的结果可以应用于不同的领域，包括安全、交通、医药、能源、制造业和许多其他领域。通过适当的修改，所提出的分布式聚类算法在网络科学与工程的各个领域也有广泛的应用。拟议研究的多学科性质，也有助于跨学科教育和全面培训未来的资讯科技人才。正在作出特别努力，鼓励来自代表性不足群体的学生参与。通过各种渠道向工业界和广大公众传播研究成果、开发的软件和教材。

ECCS-1002258Huaiyu DaiNorth Carolina State UniversityTowards a General Automatic Reasoning Framework for Networked Systems ABSTRACTIntellectual Merit: The objective of this research is to develop a general and flexible framework for automatic inference in networked systems that can admit wide applications and provide desired tradeoff between accuracy and efficiency, while allowing simple and distributed implementation. Given that exact interference is generally intractable whereas the widely-adopted mean field approach is limited in inference quality, the approach of this research is to explore the structured variational methods to bridge the gap and allow flexible tradeoff among various potentially conflicting goals, which can be tuned according to the needs and available resources of applications. Various models, dependency structures, and inference algorithms will be considered in this variational inference framework, together with efficient distributed implementation. An associated research line is to design automatic clustering schemes that ideally can match the needs of inference tasks while being easy to implement in a distributed manner. The interaction between the clustering and inference layer will be explored to achieve better performance and adaptivity. Broader Impacts: This project contributes fundamental concepts, designs, analytical tools and software towards building a general and systematic framework for automatic reasoning in large-scale networked systems, such as sensor networks and cyber-physical systems. The expected outcome can be applied in diverse areas including security, transportation, medicine, energy, manufacturing and many others. Through suitable modification, the proposed distributed clustering algorithms also admit wide applications in various areas of network science and engineering. The multi-disciplinary nature of the proposed research also lends itself to cross-disciplinary education and well-rounded training of a future IT workforce. Special efforts are taken to encourage the participation of students from underrepresented groups. Various channels are pursued to disseminate research findings, developed software, and educational materials to industry and the broader public.