Hal Cooper, Garud Iyengar , and Ching-Yung Lin, Columbia University, USA

Graph databases and distributed graph computing systems have traditionally abstracted the design and execution of algorithms by encouraging users to take the perspective of lone graph objects, like vertices and edges. In this paper, we introduce the SmartGraph, a graph database that instead relies upon thinking like a smarter device often found in real-life computer networks, the router. Unlike existing methodologies that work at the subgraph level, the SmartGraph is implemented as a network of artificially intelligent Communicating Sequential Processes. The primary goal of this design is to give each “router” a large degree of autonomy. We demonstrate how this design facilitates the formulation and solution of an optimization problem which we refer to as the “router representation problem”, wherein each router selects a beneficial graph data structure according to its individual requirements (including its local data structure, and the operations requested of it). We demonstrate a solution to the router representation problem wherein the combinatorial global optimization problem with exponential complexity is reduced to a series of linear problems locally solvable by each AI router

Intelligent Information, Database Systems, Graph Computing