Main research interests:

Trustworthy and Explainable AI

• AI systems that operate reliably, transparently, and ethically, ensuring fairness, accountability, and respect for privacy and fundamental rights. Transparent and trustworthy AI systems via explainable-by-design models and post-hoc techniques.
• Complex data analysis and decision-making in sensitive domains based on (combined) results from machine/deep learning, symbolic reasoning, and agentic AI.

Deep Image Analysis

Harnessing AI, machine learning, and computer vision to process, interpret, and optimize images across various application domains:

• Medical diagnostics (e.g., X‑rays, MRI, CT scans)
• Structural analysis in construction (e.g., automated visual inspections, thermography)
• Surveillance and security (e.g., facial recognition, video analysis)
• Industrial defect detection (e.g., quality control in manufacturing processes)
• Remote sensing and geospatial analysis (e.g., satellite images, environmental monitoring)

Agentic Multimodal Retrieval-augmented Reasoning and Generation.

Integration of modular components for data representation, semantic indexing, and knowledge retrieval across heterogeneous modalities using an agentic AI approach.

• Adaptive coordination: autonomous processing agents to enable dynamic, interpretable, and goal-oriented information synthesis.
• Coherent and contextually grounded multimodal generation through the fusion of symbolic and subsymbolic representations within an extensible computational architecture.

Green and Sustainable AI

Development of energy-efficient AI methodologies aimed at minimizing computational footprint across the full model lifecycle.

• Energy-aware model design leveraging structured pruning, mixed-precision quantization, and hardware-adaptive Neural Architecture Search (NAS) to reduce parameterization and inference cost.
• Sustainable HPC orchestration integrating dynamic power management, topology-aware resource allocation, and thermally optimized workload distribution.
• Carbon footprint assessment through standardized energy-to-solution metrics, lifecycle impact modelling, and real-time energy profiling.