Branka Stojanović

Research profile

Research summary

Branka Stojanović is a researcher at Joanneum Research in Austria whose recent work spans two applied machine-learning areas: adversarial robustness for industrial control system (ICS) security and explainable hyperspectral image classification. Her 2026 first-author preprint, 'Towards a Resilience-Theoretic Foundation for Adversarial Robustness in Industrial Control System Anomaly Detection', maps detector robustness onto resilience theory and proves a coupling-weighted system bound, with work grounded in the BATADAL setting. A second 2026 paper applies Grad-CAM to a small Conv1D network to select key SWIR wavelengths for wood-recycling hyperspectral imaging, judged plausible but short of baselines and variance reporting. Her longer-run OpenAlex topic profile sits in security and forensics: biometric identification and security (24 records), handwritten text recognition (18), network security and intrusion detection (15), forensic fingerprint detection (14), and information and cyber security (10). The snapshot records an h-index of 13, an i10-index of 16, 12 works in the last three years, and 7 works already in 2026, yielding an activity score of 75. Collaboration is comparatively narrow (score 36, five distinct recent co-authors), and reproducibility across the two reviewed papers averages 38. Output and impact axes remain unscored for lack of underlying data. This profile is name-resolved from public bibliographic data (OpenAlex; ORCID 0000-0002-5459-0507).

Research keywords

Highlights

• Resilience-theoretic foundation for ICS anomaly detection (first author, 2026)

  The first-author preprint maps the robustness of industrial control system anomaly detectors onto resilience theory and proves a coupling-weighted system bound, framed on the BATADAL benchmark. Evolve finds the direction promising but notes the bound is a weighted minimum; reproducibility readiness is rated medium.

• Grad-CAM-guided wavelength selection for wood-recycling hyperspectral imaging (2026)

  In this middle-author arXiv paper, Grad-CAM applied to a small Conv1D network selects key SWIR wavelengths for wood-recycling hyperspectral classification. Evolve judges the results plausible but flags the absence of PCA/random baselines and variance reporting; reproducibility readiness is rated low.

• A decade-scale security and forensics base under the recent pivot

  OpenAlex topic counts show biometric identification and security (24), handwritten text recognition (18), network security and intrusion detection (15), forensic fingerprint detection (14), and information and cyber security (10), alongside an h-index of 13 and i10-index of 16 — the accumulated base from which the 2026 adversarial-ML and hyperspectral XAI work departs.

Timeline

1. 2012Earliest recorded works appear in the OpenAlex counts-by-year, with 1 work recorded in 2012.
2. 2019Most productive recorded year: 14 works recorded in the OpenAlex counts-by-year.
3. 2023Recorded output falls to 2 works — a level matched again in 2025 — before the 2026 rebound.
4. 2026Two arXiv preprints enter the Evolve catalog — first-author resilience-theoretic ICS anomaly detection and middle-author hyperspectral XAI dimensionality reduction — within a rebound year of 7 recorded works.

Papers in the Evolve catalog

• 2026-09-16

  Dimensionality reduction for AI based hyperspectral image classification based on XAI

• 2026-09-07first author

  Towards a Resilience-Theoretic Foundation for Adversarial Robustness in Industrial Control System Anomaly Detection

Frequent co-authors