Talk by Milos Brajovic

19. May 2026
Title: Spectral Signal Analysis on Directed Graphs
Time: May 22, 2026 @1pm
Location: Sem. 402

Abstract: Spectral analysis on directed graphs is considerably more delicate than in the undirected case, due to non-symmetric operators, complex spectra, and the lack of a straightforward frequency interpretation. The difficulty is especially pronounced for directed acyclic graphs (DAGs), where the adjacency matrix is nilpotent, causing all eigenvalues to collapse to zero and making the standard graph Fourier transform impractical.

The presentation focuses on graph zero-padding and related spectral tools for directed graphs, with particular emphasis on DAGs. By augmenting the graph with a structured return path and zero-valued added vertices, the vertex-domain behavior of graph FIR systems is preserved while meaningful adjacency-based spectral analysis, frequency interpretation, and spectral filter design become feasible.
If time permits, matched filtering on graphs and its relation to graph convolutional neural networks will also be briefly discussed.

Our guest Miloš Brajović, is an Assistant Professor at the University of Montenegro

for a CV See:  https://www.tfsa.me/milosb_cv.html