We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: SGL: Spectral Graph Learning from Measurements

Authors: Zhuo Feng
Abstract: This work introduces a highly scalable spectral graph densification framework for learning resistor networks with linear measurements, such as node voltages and currents. We prove that given $O(\log N)$ pairs of voltage and current measurements, it is possible to recover ultra-sparse $N$-node resistor networks which can well preserve the effective resistance distances on the graph. Also, the learned graphs preserve the structural (spectral) properties of the original graph, which can potentially be leveraged in many circuit design and optimization tasks. We show that the proposed graph learning approach is equivalent to solving the classical graphical Lasso problems with Laplacian-like precision matrices. Through extensive experiments for a variety of real-world test cases, we show that the proposed approach is highly scalable for learning ultra-sparse resistor networks without sacrificing solution quality.
Comments: 2021 ACM/IEEE Design Automation Conference (DAC)
Subjects: Machine Learning (cs.LG); Social and Information Networks (cs.SI); Signal Processing (eess.SP)
Cite as: arXiv:2104.07867 [cs.LG]
  (or arXiv:2104.07867v1 [cs.LG] for this version)

Submission history

From: Zhuo Feng [view email]
[v1] Fri, 16 Apr 2021 03:01:15 GMT (5487kb,D)

Link back to: arXiv, form interface, contact.