Mathematics > Optimization and Control
[Submitted on 23 Dec 2011 (v1), revised 12 Jul 2012 (this version, v2), latest version 13 Sep 2014 (v5)]
Title:(Brain) Graph Matching via Fast Approximate Quadratic Programming
View PDFAbstract:Graph matching (GM)---the process of finding an optimal permutation of the vertices of one graph to minimize adjacency disagreements with the vertices of another---is rapidly becoming an increasingly important computational problem, arising in fields ranging from machine vision to chemical engineering to neuroscience. Because GM is NP-hard, exact algorithms are unsuitable for today's massive graphs; yet, scalable GM algorithms have received short shrift. GM can be formulated as a quadratic program with linear and binary constraints. We develop a fast approximate quadratic (FAQ) assignment algorithm to approximately solve a relaxed quadratic program with only linear constraints. FAQ scales cubicly with the number of vertices, and demonstrates marked improvements over previous state-of-the-art on nearly all benchmarks. Moreover, our non-convex formulation facilitates multiple restarts; 2-3 wisely chosen initial conditions yield the best objective function on all benchmarks. We find qualitatively similar results for our motivating application: brain-graph matching. Unfortunately, the computational complexity of FAQ scales too poorly to use it for mammalian brain-graphs, with millions or billions of vertices. To inspire further development of approximate solutions to these problems, this work is available from on the first author's website, this http URL.
Submission history
From: Joshua Vogelstein [view email][v1] Fri, 23 Dec 2011 02:56:26 UTC (54 KB)
[v2] Thu, 12 Jul 2012 03:44:51 UTC (80 KB)
[v3] Thu, 15 Nov 2012 04:56:59 UTC (133 KB)
[v4] Fri, 10 May 2013 01:58:09 UTC (193 KB)
[v5] Sat, 13 Sep 2014 15:08:28 UTC (428 KB)
Current browse context:
math.OC
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.