layout: post | title: “Learning Multi-Frequency Partial Correlation Graphs: Preprint version available.” | date: 2023-12-07 20:00:00 +0100 | categories: preprints

Check out here the preprint version of Learning Multi-Frequency Partial Correlation Graphs! It’s a joint work with my PhD Advisor and Professors from the signal processing research group of the Department of Information Engineering, Electronics, and Telecommunications (DIET) of Sapienza University of Rome.

  • Motivation: In many applications it is pivotal to discriminate partial correlations occurring at different frequency bands. State-of-the-art methods fail in this discrimination.
  • Our contribution:
    • Propose the learning of a multi-frequency Partial Correlation Graph, where different layers correspond to different frequency bands, and where partial correlations can possibly occur only over some frequency bands.
    • Formulate and solve two nonconvex learning problems to accomplish this task. Our methods do not rely on any specific statistical model.
    • We jointly learn the cross-spectral density matrices and their inverses.
  • Results:
    • Outperform baseline methods in this task.
    • Successful application on financial portfolios.

Link to the code in the supplemental.