Elvin Isufi

Research profile

Research summary

Elvin Isufi is a signal processing and graph machine learning researcher at the University at Albany, State University of New York, whose profile is name-resolved from OpenAlex bibliographic data (ORCID 0000-0002-1919-260X). The bibliometric snapshot records an h-index of 25, an i10-index of 51, and a steady output curve that climbs from single works in 2014-2015 to 16-19 works annually since 2020, with 46 works in the most recent three years and 15 works already recorded for 2026. Topic counts concentrate on Advanced Graph Neural Networks (93 mentions), followed by Neural Networks and Applications (34), Graph Theory and Algorithms (29), Topological and Geometric Data Analysis (27), and Mathematical Analysis and Transform Methods (24). Three papers reviewed by Evolve — all 2026 arXiv preprints — show a consistent agenda: Kronecker-structured spatiotemporal covariance networks for forecasting, an unoriented topological signal processing framework with a proven interaction-order decomposition, and a tutorial connecting covariance-based neural networks to PCA with stability guarantees. Evolve's verdicts praise theoretical soundness and parameter efficiency while flagging dated baselines and thin experiments; reproducibility readiness is rated high for two papers and medium for one, averaging 85 across the three. Collaboration metrics remain modest: 10 distinct coauthors, with repeat collaborations around Andrea Cavallo, Geert Leus, Gonzalo Mateos, and Alejandro Ribeiro.

Research keywords

Highlights

• Kronecker-structured spatiotemporal forecasting

  Evolve's review of KVNN finds that decoupling spatiotemporal covariance via Kronecker sums yields sound theory and parameter-efficient forecasting, though baselines are dated and margins modest.

• Unoriented operators for topological signal processing

  An unoriented topological signal processing framework with a proven interaction-order decomposition; Evolve finds the theory sound and the induced regularizers winning on four small real datasets.

• Linking covariance matrices, PCA, and graph learning

  A tutorial showing that covariance neural networks (GNNs on covariance graphs) extend PCA with stability and multiscale transferability theory; Evolve judges the claims sound but experiments illustrative and thin.

Timeline

1. 2014First works appear in the bibliometric record, with 1 work in 2014 followed by 1 in 2015.
2. 2016Output rises to 7 works, beginning a sustained growth phase that continues through the following years.
3. 2020Output reaches 16 works, up from 10 in 2019, entering a sustained regime of 16-19 works per year.
4. 2023Annual output peaks at 19 works in the snapshot, the highest single-year count on record.
5. 2026Three arXiv preprints are reviewed by Evolve — Kronecker covariance networks, unoriented topological signal processing, and a covariance/PCA tutorial — with 15 works recorded for the year.

Papers in the Evolve catalog

• 2026-09-21

  Spatiotemporal Kronecker Covariance Neural Networks

• 2026-09-21

  Topological Signal Processing With Unoriented Operators

• 2026-09-09

  Learning with Covariance Matrices: Principal Component Analysis Meets Learning with Graphs

Frequent co-authors