Balliu et al. (DISC 2020) classified the hardness of solving binary labeling problems with distributed graph algorithms; in these problems the task is to select a subset of edges in a $2$-colored tree in which white nodes of degree $d$ and black nodes of degree $δ$ have constraints on the number of selected incident edges. They showed that the deterministic round complexity of any such problem is $O_{d,δ}(1)$, $Θ_{d,δ}(\log n)$, or $Θ_{d,δ}(n)$, or the problem is unsolvable. However, their classification only addresses complexity as a function of $n$; here $O_{d,δ}$ hides constants that may depend on parameters $d$ and $δ$.
In this work we study the complexity of binary labeling problems as a function of all three parameters: $n$, $d$, and $δ$. To this end, we introduce the family of structurally simple problems, which includes, among others, all binary labeling problems in which cardinality constraints can be represented with a context-free grammar. We classify possible complexities of structurally simple problems. As our main result, we show that if the complexity of a problem falls in the broad class of $Θ_{d,δ}(\log n)$, then the complexity for each $d$ and $δ$ is always either $Θ(\log_d n)$, $Θ(\log_δn)$, or $Θ(\log n)$.
To prove our upper bounds, we introduce a new, more aggressive version of the rake-and-compress technique that benefits from high-degree nodes.
Deterministic Distributed Algorithms and Measurable Combinatorics on $Δ$-Regular Forests
CISPA Helmholtz Center for Information Security(CISPA赫尔姆霍兹信息安全中心);National University of Singapore(新加坡国立大学);University of Leipzig(莱比锡大学);ETH Zürich(苏黎世联邦理工学院);Charles University(查理大学);Eötvös Loránd University(罗兰大学)
Note on distributed certification of minimum spanning trees
分布式证明(也称为本地证明或证明标记方案)是一种证明图问题的解决方案是正确的机制。它采取的形式是将标签分配给节点,可以在本地进行检查。对于最小生成树问题存在这样的证明,使用$O(\log n \log W)$位标签(其中$n$是图中的节点的数量,并且$W$是边的最大权重)。这是由于Korman和Kutten在[Korman和Kutten 07]中以简洁和正式的方式描述了它。在这篇文章中,我们提出了一个更直观的描述结果,以及一个温和的介绍问题。
Fully Polynomial-Time Distributed Computation in Low-Treewidth Graphs
我们考虑全球性的问题,即问题,需要至少直径的时间,即使带宽不受限制。我们表明,所有考虑的问题承认有效的解决方案,在低树宽图。我们所说的"有效“是指运行时间对树宽有多项式依赖性,对直径有线性依赖性(这是不可避免的),对图中节点数$n$只有多对数依赖性。我们提出了解决CONGEST模型中以下问题的算法,它们都达到了$\tilde{O(τ^{O(1)}D)}$ -轮复杂度(其中$τ$和$D$分别表示图的树宽和直径):(1)精确单源最短路径(实际上,计算距离标记方案的更一般的问题),(2)精确二分未加权最大匹配,(3)有向图和无向图的加权围长。我们得出我们所有的结果使用一个统一的框架,其中包括两个新的技术成分,第一个是一个完全多项式时间的分布式树分解算法,它输出分解的宽度$O(τ^2\log n)$在$\tilde{O}(τ^{O(1)}D)$轮(其中$n$是图中的节点数)。第二个成分,也是本文的技术亮点,是\emph{stateful walk constraint}的新概念,它基于它们的局部属性(例如,增强路径)。给定有状态的行走约束,最短路径问题(或距离标记)的约束版本要求算法输出给定源顶点和宿顶点的最短\emph{constrained}行走(或其距离)。我们表明,这个问题可以有效地解决在CONGEST模型,减少它的\emph{unconstrained}版本的问题。
Short Labeling Schemes for Topology Recognition in Wireless Tree Networks
我们认为在无线(无线电)网络建模为无向图的拓扑识别问题。拓扑识别是一项基本任务,其中网络的每个节点都必须输出底层图的映射,即,它的一个同构副本,并在这张地图上显示它自己。在无线网络中,节点在同步轮中通信。在每一轮中,节点可以向所有邻居发送消息,也可以保持沉默并进行监听。在接收端,如果$v$在给定轮中监听,并且如果$w$是其在该轮中发送的唯一邻居,则节点$v$在该轮中监听来自邻居$w$的消息。节点的标签是二进制字符串(不一定不同)。标签方案的长度是标签的最大长度。我们专注于无线网络建模树,我们调查两个问题。\begin{itemize} \item What is the shortest labeling scheme that permits topology recognition in all wireless tree networks of diameter $D$ and maximum degree $Δ$? \item What is the fastest topology recognition algorithm working for all wireless tree networks of diameter $D$ and maximum degree $Δ$, using such a short labeling scheme? \end{itemi…
Sparse Binary Compression: Towards Distributed Deep Learning with minimal Communication
Fraunhofer Heinrich Hertz Institute(弗劳恩霍夫海因里希·赫兹研究所);Technische Universität Berlin(柏林工业大学);Max Planck Institute for Informatics(马克斯·普朗克信息学研究所);Korea University(高丽大学)
Binary Bleed: Fast Distributed and Parallel Method for Automatic Model Selection
Theoretical Division, Los Alamos National Laboratory(洛斯阿拉莫斯国家实验室理论部);Advanced Research in Cyber Systems, Los Alamos National Laboratory(洛斯阿拉莫斯国家实验室网络系统高级研究部);Department of Computer Science and Electrical Engineering, University of Maryland, Baltimore County(马里兰大学巴尔的摩分校计算机科学与电气工程系)