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

Download:

Current browse context:

math.NA

Change to browse by:

References & Citations

Bookmark

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

Mathematics > Numerical Analysis

Title: pISTA: preconditioned Iterative Soft Thresholding Algorithm for Graphical Lasso

Abstract: We propose a novel quasi-Newton method for solving the sparse inverse covariance estimation problem also known as the graphical least absolute shrinkage and selection operator (GLASSO). This problem is often solved using a second-order quadratic approximation. However, in such algorithms the Hessian term is complex and computationally expensive to handle. Therefore, our method uses the inverse of the Hessian as a preconditioner to simplify and approximate the quadratic element at the cost of a more complex \(\ell_1\) element. The variables of the resulting preconditioned problem are coupled only by the \(\ell_1\) sub-derivative of each other, which can be guessed with minimal cost using the gradient itself, allowing the algorithm to be parallelized and implemented efficiently on GPU hardware accelerators. Numerical results on synthetic and real data demonstrate that our method is competitive with other state-of-the-art approaches.
Subjects: Numerical Analysis (math.NA); Mathematical Software (cs.MS)
Cite as: arXiv:2205.10027 [math.NA]
  (or arXiv:2205.10027v1 [math.NA] for this version)

Submission history

From: Eran Treister [view email]
[v1] Fri, 20 May 2022 08:48:26 GMT (77kb,D)

Link back to: arXiv, form interface, contact.