AF: Small: Graphs and structures for distance estimation

网络中的距离计算与估计是网络分析中最基本、最基本的任务之一。然而，存储图中每对节点之间的距离是不可行的，特别是对于今天的“大数据”。一个图的小草图已经被开发出来，这样任何两两距离的一个很好的估计都可以从草图中得到。这个项目旨在推动这种草图的艺术水平。主要研究对象是扳手、距离预言器及其容错变体。扳手是一种稀疏子图，它不会将原始图的任何距离拉伸太多。距离oracle是一种空间占用很小的数据结构，能够有效地回答（近似）距离查询。spanner和distance oracle都压缩距离信息。它们之间的主要区别在于，人们仍然可以在扳手上运行图形算法，但不能在距离oracle上运行，而人们可以立即从距离oracle中获得任何距离，但在扳手中，人们必须实际计算它。在实践中，网络本质上是动态的。为了解决这个问题，图形草图必须容忍错误，即边和顶点的删除。有容错版本的扳手和距离预言器——这些结构估计任何给定的（通常是固定大小的）故障边或节点子集的距离。该项目将提供构建新的低空间扳手和oracle的算法，并提供改进的保证，并将努力开发容错和距离估计的新技术。spanner和distance oracles有许多应用，例如用于距离计算的并行和分布式算法，网络路由，以及在非同步网络中模拟同步协议。更好地理解沿短路径的网络路由可以指导下一代互联网协议的设计。该研究还与度量嵌入领域密切相关，因此超出了计算机科学的严格界限。这项研究的材料将被整合到核心的本科和研究生课程中，并将导致有关该主题的新课程的开发。课堂讲稿和项目材料将在课程网站上提供给公众。

Distance computation and estimation in networks is one of the most basic and fundamental tasks in network analysis. However, storing the distance between every pair of nodes of a graph is infeasible, especially for today's "big data." Small sketches of a graph have been developed so that a good estimate of any pairwise distance can be retrieved from the sketch. This project aims to advance the state-of-the art of such sketches. The main objects of study are spanners, distance oracles and their fault-tolerant variants. A spanner is a sparse subgraph that does not stretch any distance of the original graph by much.  A distance oracle is a data structure that has small space usage and is capable of answering (approximate) distance queries efficiently. Both spanners and distance oracles compress the distance information. The main difference between them is that one can still run graph algorithms on a spanner, but not on a distance oracle, whereas one can obtain any distance from a distance oracle instantly, but in a spanner one would have to actually compute it.In practice, networks are dynamic in nature. To address this, graph sketches would have to tolerate faults, i.e. edge and vertex deletions. There are fault-tolerant versions of spanners and distance oracles-- these structures estimate distances for any given (typically fixed size) subset of failed edges or nodes. This project will provide algorithms for constructing new low-space spanners and oracles with improved guarantees and will strive to develop new techniques for fault-tolerance and distance estimation in general. Spanners and distance oracles have many applications, e.g. in parallel and distributed algorithms for distance computation, network routing, and simulating synchronized protocols in unsynchronized networks. A better understanding of network routing along short paths could guide the design of next-generation Internet protocols. The research is also closely tied to the field of metric embedding, and thus extends beyond the strict boundaries of computer science. Material from this research will be integrated into core undergraduate and graduate courses, and will lead to the development of new courses on the topic. The lecture notes and project materials will be available on the course website for the general public.