III: Small: Temporal Relational Triples, or TR2: A Novel Data and Knowledge System for Temporal and Streaming Data

点击翻译按钮获取中文摘要

This project extends a multiple-data-stream system with the state-of-the-art relational machine learning, and achieves a holistic data and knowledge system for comprehensive querying, interpretation, inference, event monitoring, and actions. In a real application scenario, temporal and streaming data typically come from multiple sources including mobile, IoT devices and sensors, system logs, Internet data, high-volume feeds from social networks, among others. Different data sources may have data in different formats, structures, and schema, and a significant challenge in integrating such diverse sources of data streams is the interoperability. This project will devise an approach called temporal relational triples (TR2), which extends triple stores that have unique strengths in data integration and interoperability with coding storage and knowledge components. TR2 is a pioneering attempt to bridge the gap between neural embedding based modern AI that has proven to be effective and comprehensive and interpretable analytics of temporal and streaming data. The results will impact domains that make use of temporal and streaming data, such as healthcare, banking and finance, cybersecurity, social network analysis, mobile and pervasive computing, and sensor monitoring for various purposes.TR2 aims to advance the frontier of data stream and temporal data management systems by leveraging modern relational machine learning and representation learning. It is also a generalization and extension of previous work in triple stores, functional dependency, active databases, complex event processing, predictive queries, and rule/knowledge systems, in a novel holistic data system, all tightly woven through the threads of coding and relational machine learning. TR2 incorporates the following four key new ideas: (1) flexible integration of streams from multiple sources as temporal relational triples, (2) introducing coding using relational machine learning as a major component of TR2, and extending data with predicted counterparts for comprehensive queries, (3) learning higher-level rules and knowledge with coding, which are useful for interpretation and probabilistic inference, and (4) extracting higher-level states and patterns for continuously monitoring the imminence of critical events and for choosing actions. The project ingrains neural network embedding into a universal treatment of entity-relationship view of data and information model central to data management.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.