SPECTRAL GRAPH THEORETIC METHODS FOR ESTIMATION AND CONTROL
ROBOTICS: SCIENCE AND SYSTEMS
FRIDAY, JULY 14, 2023
DAEGU, REPUBLIC OF KOREA
Room 324B and accessible virtually on PheedLoop
Thank you for attending!
Thank you to all of our speakers and attendees for making the RSS 2023 Workshop on Spectral Graph-Theoretic Methods for Estimation and Control (SGTM 2023) a resounding success! We had a great time, and we hope we can revisit this rich and rewarding topic again at a future conference with you all.
Videos of all our our speakers' recorded talks can now be found on YouTube! Please see the playlist below for all presentations.
OVERVIEW
Graphical structure is ubiquitous in robotics, computer vision, and machine learning applications, including, for example, probabilistic inference, multi-agent estimation and control, and distributed learning. Consequently, understanding what specific features of graphs influence the properties of problems defined over them (and how), provides a powerful unifying lens for studying a wide range of practically-important problems, as well as devising new algorithmic approaches to solve them. In particular, recent work in robotics and computer vision has shown algebraic graph theory to be especially useful. In brief, this field of mathematics studies properties of graphs by means of algebraic objects (e.g., matrices and their spectra, linear spaces, and/or groups) constructed from them, thus bringing to bear the powerful tools of algebra. These techniques have recently enabled sharp novel insights into the intrinsic difficulty of several fundamental problems in estimation and control, as well as novel state-of-the-art algorithms.
The goal of this workshop is to introduce and promote this powerful suite of tools to a broader robotics audience. In particular, the technical program will be structured around the following three themes:
Mathematical fundamentals: What is algebraic graph theory, and what is it good for?
Current research: How are researchers currently using these techniques to devise state-of-the-art algorithms in robotics, computer vision, and machine learning?
Future directions: What kinds of emerging applications and new research frontiers might these techniques enable in the future?
EXPECTED OUTCOMES
Foundations: What is algebraic graph theory, and what is it good for?
Applications: What are the latest results and emerging applications for algebraic/spectral graph theory in robotics? What open challenges are there toward improving the application of these techniques?
Frontiers: What are emerging capabilities that people have not yet exploited for these kinds of problems?
INVITED SPEAKERS
Federica Arrigoni
Politecnico di Milano, Italy
Luana Ruiz
MIT, USA
Yongbo Chen
Australian National University, Australia
Julio Placed
University of Zaragoza, Spain
Yulun Tian
MIT, USA
Kevin Doherty
MIT, USA
SCHEDULE
13:30 - 15:00 Session 1:
13:30 - 14:00 Welcome and Introduction
14:00 - 14:30 Luana Ruiz, Large-scale graph machine learning: tradeoffs, guarantees and dynamics
14:30 - 15:00 Federica Arrigoni, Synchronization Problems in Computer Vision
15:00 - 15:30 Coffee break
15:30 - 17:30 Session 2:
15:30 - 15:55 Kevin Doherty, MAC: Maximizing Algebraic Connectivity for Graph Sparsification
15:55 - 16:20 Yulun Tian, Rotation Averaging via Fast Laplacian Solvers
16:20 - 16:45 Julio Placed, Exploiting the Graph Spectrum to Make Optimal Decisions in Active SLAM
16:45 - 17:10 Yongbo Chen, Spectral Graph-Theoretic Methods for SLAM
CALL FOR SUBMISSIONS
We welcome extended abstracts (up to 1 page) on new applications of algebraic and/or spectral graph theory (broadly construed) in robotics, computer vision, control engineering, and machine learning. Submissions will be reviewed by the organizers and a program committee and those accepted will be featured in lightning talks and posters.
Abstract Submission: https://forms.gle/7vUSAvmCuJYcmoKq9
Submission Deadline: Friday, June 23, 2023 - 11:59 PM AoE
ORGANIZERS
David Rosen
Northeastern University
Kevin Doherty
MIT
Kasra Khosoussi
CSIRO
Matt Giamou
Northeastern University