SGER: Algorithmic Infrastructure for Knowledge Management

这个项目探索了复杂知识的存档、索引和再利用的算法技术。层次图已经成为一种强大的知识表示结构，然而现有的数据库和数据挖掘技术不能处理将这些结构作为原始数据元素进行比较、分类、索引和聚类所需的极端组合复杂性。研究挑战包括：(1)获取基于图的模型的内部结构及其拓扑子结构，以用于模式匹配；(2)分类这些近似技术在存在噪声的情况下的稳定性；(3)确定如何将这些技术转换为合适的数据库机制。该方法的基础是通过特征值刻画将图的拓扑结构映射到低维向量空间。该SGER在协作和分布式工程设计的背景下研究这些问题，其中代理团队(人和计算)通过网络交互以实现产品(例如，软件、机电设备、建筑等)。这个过程被建模为一个有向无环图(DAG)，它编码了设计建模操作和决策，以及沿着建模时间步骤的信息流。将派出知识获取代理来获取这些知识结构，并将对工艺知识的索引和再利用的理论方法进行验证。其中一些工作将与宾利系统公司合作完成。如果成功，这些技术将为数据库和信息管理系统带来戏剧性的新可能性：允许它们将复杂的图形有效地存储为单个大型对象，并在这些组合结构的大集合中识别有用的模式。

This project explores algorithmic techniques for archival, indexing and reuse of complex knowledge. Hierarchical graphs have emerged as a powerful knowledge representation structures, however existing database and data mining techniques are not able to manage the extreme combinatorial complexities required to compare, classify, index and cluster these structures as primitive data elements. Research challenges include (1) capturing the inner structure of graph based models, and their topological sub-structures, to use for pattern matching purposes; (2) classify the stability of these approximation techniques in the presence of noise; (3) identifying how to translate these techniques into suitable database mechanisms. The basis of the approach is a mapping of the topological structure of a graph into a low-dimensional vector space through an eigenvalue characterization. This SGER studies these problems in the context of collaborative and distributed engineering design, where teams of agents (human and computational) interact over the network to realize a product (e.g., software, electro-mechanical device, building, etc.). This process is modeled as a directed acyclic graph (DAG) that encodes the design modeling operations and decisions, as well as the flow of the information along the modeling time-steps. Knowledge acquisition agents will be fielded to capture these knowledge structures and a validation of the theoretical methodology for indexing and reuse of process knowledge will performed. Some of this work will be done in collaboration with Bentley Systems. If successful, these techniques will lead to dramatic new possibilities for database and information management systems: allowing them to efficiently store complex graphs as single large objects and identify useful patterns in and across large sets of these combinatorial structures.