Real-Time Probabilistic Inference

这项研究的目标是扩展贝叶斯推理的计算限制，专注于嵌入式、实时问题。贝叶斯概率既是一种理论框架，也是一种代表性和计算性方法，对人工智能的研究和实践产生了深远的影响。然而，现有计算方法的局限性仍然阻碍了它的广泛应用，特别是在实时和嵌入式系统中。这项研究以基于搜索的近似方法为核心，使用实时决策评估作为其实验工具，并扩展到分布式系统中的可伸缩问题，包括编译(包括强化学习)和通信控制。这项研究的结果将包括工业、医疗和消费应用的实时监测和评估的改进方法。

The goal of this research is to stretch the computational limits of Bayesian inference with focus on embedded, real-time problems. Bayesian probability, as both a theoretical framework and as a representational and computational method, is having a profound impact on artificial intelligence research and practice. However, limitations is available computational methods still hamper widespread application, especially in real- time and embedded systems. This research builds from a core focus on search-based approximation methods, using real-time decision evaluation as its experimental tool, and extends to scaling issues including compilation (including reinforcement-learning) and control of communication in distributed systems. Results of this research will include improved methods for real-time monitoring and assessment for industrial, medical and consumer applications.