Dipartimento di Ingegneria informatica, automatica e gestionale

• Multimedia Forensics: Detection and attribution of manipulated and AI-generated media, spanning deepfake video/audio detection, synthetic image attribution, and spoofing/anti-fraud analysis.

• Adversarial & Robustness: Study of adversarial vulnerabilities in ML pipelines, including data poisoning attacks and robustness evaluation of forensic detectors under adversarial conditions.

• Explainable AI & Interpretability: Development of interpretability tools to explain the decisions of deep models, applied to deepfake detectors and depth estimation networks.

2. Lightweight and Efficiency

• Real-world applications: Deployment of efficient and interpretable vision models for domain-specific tasks such as precision sericulture (bridging the reality gap in silkworm rearing via interpretable feature streams) and urban traffic/mobility monitoring. 

• Embodied AI: Perception models for agents interacting with the physical world, including pedestrian intention estimation and efficient spatio-temporal predictive learning.

• VR/AR: Real-time and lightweight solutions for immersive environments, such as efficient 3D object reconstruction for virtual content.

3. Computer Vision Applications

• Earth Observation: Ground-to-aerial image matching, semantic segmentation, and generative (diffusion-based) methods applied to satellite/aerial imagery for tasks like change detection and misinformation verification.

• Computer Vision Tasks: Core vision problems including monocular depth estimation, optical flow, video generation, and 3D scene reconstruction.

• Medical Imaging: AI-assisted diagnostic support, including dental radiograph analysis for automated tooth segmentation and periapical lesion detection.

4. Edge Computing

• Edge AI & TinyML: Design of efficient CNN architectures optimized for deployment on resource-constrained embedded hardware.

• Computational Offloading & Task Scheduling: Algorithms for dynamic resource allocation, container autoscaling, and reinforcement-learning-based scheduling across fog/edge computing infrastructures.

• Human Computer Interaction: Design and evaluation of intelligent XR interfaces to support human decision-making in safety-critical scenarios.