AI4Industry Legal and Financial (AI4ILF)

The AI4Industry Legal and Financial (AI4ILF) is a cutting-edge research area dedicated to the development and application of bio-inspired models for deep learning in industrial, legal, and financial/commercial domains. The reaserch area’s mission is to advance artificial intelligence (AI) methodologies by leveraging bio-inspired computational models, innovative mathematical approaches, and novel learning and testing procedures to build AI-driven solutions.

In the industrial field, the research focuses on automotive applications (e.g., driver assistance, autonomous driving, engine control, powertrain optimization) and semiconductor technologies (e.g., production line optimization, semiconductor modeling, CFD analysis, AI-enhanced SPICE models, residual lifetime prediction etc...). Concurrently, the research area explores legal and financial AI applications, with a special emphasis on civil, corporate, and contract law, as well as financial risk assessment, insolvency prediction, AI-based company valuation, and advanced time-series forecasting.

The research extends across advanced AI paradigms, including convolutional neural networks (CNNs), transformers, hybrid systems, reinforcement learning (RL), and innovative learning architectures. Key topics of interest include, but are not limited to:

Perceptive deep learning for industrial applications;
Adversarial methods, regularization techniques, and adaptive Jacobian regularization;
Continual learning and mathematical analysis of learning dynamics, addressing catastrophic forgetting via Lipschitz properties, Extended Elastic Weight Consolidation (EWC), and bio-inspired models;
Learning in non-conventional spaces and geometries, particularly hyperbolic spaces;
Development frameworks based on perceptive deep learning and generative models;
Generative AI for legal applications;
Domain adaptation for automotive applications, including multi-scaled gradient reversal layers and generative layers for domain adaptation;
Explainability through foundation models and Cellular Neural Networks (CNNs);
Knowledge distillation for resource-constrained AI systems;
Neuro-modulation paradigms balancing stability and plasticity in deep networks;
Document analysis using generative AI within RAG attention-based architectures;
Hybrid bio-inspired models for financial data analysis;
Temporal Fusion Transformer (TFT) with TCN enhancement for robust time-series forecasting;
Attention-based models and explainability techniques for document analysis in civil and corporate law.

The AI4ILF research area is actively fostering strong scientific collaborations with renowned industry leaders, bridging fundamental AI research with real-world applications. Through these partnerships, the involved researchers aim to drive applied R&D, supporting the professional development of Ph.D. candidates, Ph.D students and so on.

Deep Learning in Un-conventional Geometric Spaces

We design learning algorithms that operate natively on non-Euclidean domains—curved manifolds (spherical, hyperbolic, toroidal, product manifolds), Lie groups (e.g., SE(3), SO(3)), graphs/simplicial complexes, and other structured topologies where Euclidean assumptions fail. Our goal is to build models whose representations, inductive biases, and optimization respect the intrinsic geometry and symmetries of the data.

Riemannian learning and optimization: geodesic layers, exp/log maps, retractions, parallel transport, and Laplace–Beltrami operators for diffusion, spectral, and PDE-aware networks.
Equivariant and curvature-aware architectures: group-equivariant GNNs/transformers, spherical/hyperbolic embeddings, and attention on general metric spaces.
Generative models on manifolds: normalizing flows and diffusion models with correct Riemannian volume terms and tractable Jacobians.
Stability and robustness: Lipschitz-controlled training, spectral constraints, and Lyapunov-inspired criteria; domain adaptation across geometry shifts.

Applications span inverse problems and scientific computing, geometric dynamics, molecular and materials modeling, geospatial/relational analytics, and device/semiconductor modeling where state spaces are naturally non-Euclidean. By unifying geometric principles with modern deep learning, we aim to deliver models that are theoretically grounded, numerically stable, and data-efficient, enabling accurate inference and generation directly on the spaces where the phenomena truly live.

Deep Learning 4 Industry

This research area focuses on developing artificial intelligence methods tailored to industrial processes, with the goal of improving efficiency, safety, and sustainability. By integrating machine learning, computer vision, and predictive analytics into production pipelines, we enable data-driven decision-making and real-time optimization of operations.

Smart manufacturing and Industry 4.0: adaptive control systems, digital twins, and process optimization through reinforcement learning and advanced forecasting models.
Predictive maintenance: anomaly detection and failure prediction for machinery and critical infrastructure, reducing downtime and extending equipment lifetime.
Robotics and automation: AI-driven robotic arms and autonomous systems for assembly, inspection, and quality control in complex production lines.
Energy-efficient operations: optimization of energy consumption across industrial plants, from renewable integration to demand-side management.
Human–AI collaboration: safe, explainable AI systems that enhance human operators’ decision-making while ensuring transparency, accountability, and resilience.

Applications span manufacturing, automotive, aerospace, energy, logistics, and semiconductor production, where AI enables industries to transition toward smarter, more autonomous, and sustainable ecosystems.

Deep Learning for Legal Applications

This research area explores how deep learning—both classical discriminative models and generative AI—can transform the legal domain by enabling automation, improved decision support, and enhanced access to justice.

Legal text understanding and retrieval: transformer-based models (BERT, GPT-style architectures) for case law search, contract analysis, and semantic similarity between legal documents.
Predictive legal analytics: deep learning classifiers and sequence models for outcome prediction, risk assessment, and compliance monitoring.
Generative AI for legal drafting: large language models and diffusion-based architectures to synthesize structured contracts, summarize legislation, and generate domain-specific legal arguments, while ensuring consistency and reducing redundancy.
Explainability and fairness: interpretable deep networks that respect legal standards of transparency, bias mitigation, and accountability.
Multi-modal legal intelligence: integration of text, structured case data, and evidence (e.g., images, financial traces, or blockchain ledgers) for enriched reasoning.

By combining classical deep learning with state-of-the-art generative approaches, we aim to build AI systems that support lawyers, regulators, and organizations with tools for faster information processing, more accurate predictions, and reliable drafting assistance, all while adhering to the strict ethical and normative requirements of the legal field.

Deep Learning for Financial Applications

This research area focuses on the application of artificial intelligence and deep learning to the financial sector, addressing challenges in banking, asset management, and trading. The goal is to create robust, explainable, and regulation-aware AI systems that support decision-making, risk management, and financial innovation.

Banking and credit systems: AI models for credit scoring, risk assessment, anti-money laundering (AML), fraud detection, and customer behavior prediction, enabling more secure and personalized financial services.
Trading and asset management: deep reinforcement learning and predictive modeling for algorithmic trading, portfolio optimization, and dynamic hedging strategies under non-stationary market conditions.
Financial forecasting: advanced time-series and generative models for volatility prediction, scenario simulation, and stress testing, improving resilience and strategic planning.
Regulatory technology (RegTech): AI-powered compliance monitoring, explainable models for decision traceability, and tools to support auditing and legal validation of financial processes.
Sustainable and ESG finance: data-driven analysis of environmental, social, and governance indicators to guide responsible investment strategies.

By combining classical predictive deep learning with generative and reinforcement-based approaches, we aim to develop AI frameworks that empower banks, regulators, and traders with data-efficient, adaptive, and trustworthy tools for the evolving financial ecosystem.

(2025)
          1.  F. Rundo, A. A. Messina, M. Fiore, M. Calabretta, P. Coscia and S. Battiato, "Hyperbolic Deep Learning System for Intelligent Gate-Driving and Health Monitoring of Silicon-Carbide Power MOSFETs," in IEEE Open Journal of the Computer Society, doi: 10.1109/OJCS.2025.3614091. 
          2.  F. Rundo, G. Castagnolo, C. Pino, M. Calabretta, A. Messina, S. Battiato, "'Attention-Enhanced Convolutional Deep Architecture with Adaptive Jacobian Regularization for Joint Multi-Modal Optical and X-Ray Screening of Silicon-Carbide Power Devices'" , IEEE Proceedings of the International Joint Conference on Neural Networks, IJCNN 2025, Rome - June 30 to July 5, 2025 
          3.	G. Castagnolo, C. Pino, M. Calabretta, A. Messina, S. Patane, F. Garesci, M. D'Ambrosio, S. Battiato, F. Rundo, "'DISTILLO Framework: a Novel Knowledge Distillation Platform for Advanced Intelligent Solution Deployment over Automotive-Grade Devices'" , IEEE Proceedings of the International Workshop on Metrology for Automotive , PARMA - ITALY / June 25 - 27, 2025 
          4.	M. D'Ambrosio,C. Tripodi, F. Garesci,  D. Cosio, D. Bonanno, A. Messina, A. Imbruglia, M. Calabretta, S. Patane, F. Rundo, "''Investigation into the Aging Mechanisms of a SiC-Based Power MOSFET by Thermal and Thermomechanical Analysis'" , IEEE Proceedings of the International Workshop on Metrology for Automotive , PARMA - ITALY / June 25 - 27, 2025
          5.	F. Rundo, G. Castagnolo, C. Pino, M. O. Spata, M. Calabretta, A. Messina, S. Battiato, "Silicon-Carbide Power Device X-Ray Screening via Attention-Based Deep Network for a Robust Traction-Inverter in Electric Vehicles' , IEEE Proceedings of the International Workshop on Metrology for Automotive , PARMA - ITALY / June 25 - 27, 2025  
          6.	F. Rundo, G. Castagnolo, C. Pino, M. O. Spata, M. Calabretta, A. Messina, S. Battiato, "Advanced Knowledge Distillation Pipeline with Dynamic Weighted-Smoothed Loss for a Robust Deep Network Deployment in Resource-Constrained Embedded Systemstric Vehicles' , IEEE Proceedings of the IEEE INTERNATIONAL CONFERENCE ON Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering, ANCONA, ITALY / OCTOBER 22-24, 2025  (To appear)
          7.	F. Rundo, G. Castagnolo, C. Pino, M. O. Spata, M. Calabretta, A. Messina, S. Battiato,  "Advanced Non-Local Pipeline for Domain-Adapted Intelligent Power Device Defects Analysis for a Robust Traction Inverter in Next Generation Electric Cars' , IEEE Proceedings of the IEEE INTERNATIONAL CONFERENCE ON Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering, ANCONA, ITALY / OCTOBER 22-24, 2025  (To appear)
          8.	F, Rundo, M. O. Spata, C. Pino, S. Battiato, "Dopamine-Inspired Neuro-Modulated Radiomic Pipeline for Computed-Tomography-Driven Prediction of Immunotherapy Response in Metastatic Cancer Treatment' , IEEE CVF/Proceedings of the "PHAROS - Adaptation, Fairness, Explainability in AI Medical Imaging (PHAROS-AFE-AIMI) and Competition" Workshop in conjunction with the International Conference on Computer Vision (ICCV), 2025 in Hawaii, USA, 19 – 23 October, 2025  (To appear)
          9.	A. Rondinella, F. Guarnera, F. Rundo, S. Battiato "DeepSeg-MS: A 3D network based on hierarchical multi-scale learning for MRI multiple sclerosis lesion segmentation" , IEEE CVF/Proceedings of the "PHAROS - Adaptation, Fairness, Explainability in AI Medical Imaging (PHAROS-AFE-AIMI) and Competition" Workshop in conjunction with the International Conference on Computer Vision (ICCV), 2025 in Hawaii, USA, 19 – 23 October, 2025  (To appear)

(2024)
* F. Rundo, S. Coffa, et al Sistemi e procedimenti di elaborazione di rete neurale artificiale, IT Patent Pending Nr. 102024000008095; 11 Aprile 2024;
* Spata, M.O., Battiato, S., Ortis, A., Rundo, F., Calabretta, M., Pino, C., Messina, A. Deep Learning Algorithm for Advanced Level-3 Inverse-Modeling of Silicon-Carbide Power MOSFET Devices,(2024) Proceedings of SPIE - The International Society for Optical Engineering, 12973, art. no. 1297309

(2023)
* Rundo, F. et al, Intelligent Traction Inverter in Next Generation Electric Vehicles: The Health Monitoring of Silicon-Carbide Power Modules, (2023) IEEE Transactions on Intelligent Vehicles, 8 (12), pp. 4734-4753.  DOI: 10.1109/TIV.2023.3294726
* Donida Labati, R., Piuri, V., Rundo, et al. MultiCardioNet: Interoperability between ECG and PPG biometrics (2023) Pattern Recognition Letters, 175, pp. 1-7.  DOI: 10.1016/j.patrec.2023.09.009
* Pino, C., Rundo, F., Spampinato, C., Battiato, S. An AI Architecture for Power MOS Rds-on-Driven Lifetime Estimation,(2023) 2023 Smart Systems Integration Conference and Exhibition, SSI 2023, 
* Rundo, F., Calabretta, M., Pino, C., Coffa, S., Messina, A., Spampinato, C., Battiato, S.,Deep Learning for Automatic Wafer Monitoring System, (2023) 2023 Smart Systems Integration Conference and Exhibition, SSI 2023

(2022)
* Sarpietro, R.E., Pino, C., Coffa, S., Messina, A., Palazzo, S., Battiato, S., Spampinato, C., Rundo, F. "Explainable Deep Learning System for Advanced Silicon and Silicon Carbide Electrical Wafer Defect Map Assessment", (2022) IEEE Access, 10, pp. 99102-99128. DOI: 10.1109/ACCESS.2022.3204278
* Rundo, F., Messina, et al. Deep Learning Car Driver Motion Magnified Saccadic Eye Movements for Advanced Driving Assistance System, (2022) Proceedings of the International Joint Conference on Neural Networks, 2022-July

(2021)
* F. Rundo, et al,  Procedimento di Elaborazione di Segnali Indicativi di un Livello di Attenzione di un Individuo umano, Corrispondenti Sistema, Veicolo e Prodotto Informatico, IT Patent Pending Nr. 102021000008915; 09 Aprile 2021
* Rundo, F., Conoci, S., Spampinato, C., Leotta, R., Trenta, F., Battiato, S. "Deep Neuro-Vision Embedded Architecture for Safety Assessment in Perceptive Advanced Driver Assistance Systems: The Pedestrian Tracking System Use-Case", (2021) Frontiers in Neuroinformatics, 15, art. no. 667008. DOI: 10.3389/fninf.2021.667008
* Palazzo, S., Murabito, F., Pino, C., Rundo,F. et al. "Exploiting structured high-level knowledge for domain-specific visual classification", Elsevier Pattern Recognition, Volume 112, 2021, 107806, ISSN 0031-3203, https://doi.org/10.1016/j.patcog.2020.107806
* Rundo, F., Trenta, F., Leotta, R., Spampinato, C., Piuri, V., Genovese, A., Scotti, F., Battiato, S. Advanced Temporal Dilated Convolutional Neural Network for a Robust Car Driver Identification,  (2021) Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 12668 LNCS, pp. 184-199.
* Rundo, F., Leotta, R., Piuri, V., Genovese, A., Scotti, F., Battiato, S. Intelligent road surface deep embedded classifier for an efficient physio-based car driver assistance, (2021) ICAS 2021 - 2021 IEEE International Conference on Autonomous Systems, Proceedings

(2019)
* F. Rundo, et al, Procedimento di elaborazione di un segnale elettrofisiologico, sistema, prodotto informatico e veicolo corrispondenti, IT Patent Pending Nr. 102019000005868, 16 Aprile 2019;
* F. Rundo, F. Trenta S. Conoci, S. Battiato, Procedimento di Elaborazione di Immagini, Sistema, Veicolo e Prodotto Informatico Corrispondenti, IT Patent Pending Nr. 102019000000133; 07 Gennaio 2019;
* Trenta, F., Conoci, S., Rundo, F., Battiato, S. Advanced motion-tracking system with multi-layers deep learning framework for innovative car-driver drowsiness monitorin (2019) Proceedings - 14th IEEE International Conference on Automatic Face and Gesture Recognition, FG 2019, art. no. 8756566.

(2015)
* F. Rundo, G. Di Giore, S. Battiato, A. Ortis , Procedimenti e Sistemi per Il Trattamento di Immagini Stereoscopiche, prodotti Informatici e Supporto relativi / Methods And Systems for the Treatment Of Stereoscopic Images,Corresponding Computer Program, Products, And Storage Carrier, IT Patent Pending TO2012A000647 del 24.07.2012 ; USA Patent Pending US 9,183,644 B2 del 10.11.2015

Prof. Sebastiano Battiato — sebastiano.battiato@unict.it

Prof. Francesco Rundo — francesco.rundo@unict.it

Sebastiano Battiato [Contact Person]	Francesco Rundo [Responsabile Scientifico / Principal Investigator][Contact Person]

Massimo Orazio Spata
[Researcher]

Rosa Zuccarà [PhD Student]	Dario Pishvai [PhD Student]	Ignazio Zangara [PhD Student]

Kristian Pio Pintaudi [Student]	Francesco Maria Pennisi [Student]	Emanuele Rosario Fiorito [Student]	Adriano Bonanno [Student]

(2025)

First-stage passed RESEARCH PROJECT(s):

4. HORIZON-JU-Chips-IA: FA4EU (Submission Pending).
5. HORIZON-JU-Chips-RIA: CIRCLE (Submission Pending).

Next Generation of edge AI crossing technology fields (NeAIxT) - HORIZON-JU-Chips-2024-1-IA-T1 - Grant Agreement - GAP-101194172

Europe is poised to play a pivotal role in the technological revolution, particularly in the field of edge AI, which promises sustainable growth, performance, and reliability. The NeAIxt project is a strategic initiative designed to foster European independence and control over edge AI technology, benefiting both companies and citizens. The project presents a golden opportunity for European SMEs to grow, network, and enhance skills, leveraging exposure to the global market. Research labs and RTOs will bridge the gap to the future by developing necessary technologies and competencies, while universities will cultivate and disseminate advanced skills required for this technological evolution. NeAIxt aims to solidify Europe's position in edge AI and eNVM technology by enhancing AI enablers, evolving eNVM for edge applications, and demonstrating AI capabilities at both chip and system levels. The project is committed to ensuring the safety and security of AI services, adhering to EU regulations. Key technical developments include the advancement of 18nm FD-SOI and next-generation embedded Phase Change Memory (ePCM). These innovations will lead to high-performance, secure microcontrollers with AI capabilities, offering low power consumption and high security for smart applications. The project will integrate advances in non-volatile memory technologies with cutting-edge MCU design to enable efficient in-memory computing. This synergy will deliver a fully European solution for reliable, safe, and independent edge AI applications. NeAIxt will address the entire edge AI value chain, from academia to industry, and from design to end-user applications, building on Europe's strong technological foundation. The project's outcomes will alleviate societal concerns about AI proliferation by ensuring compliance with European privacy standards, fostering AI adoption in various sectors.

HICONNECTS (Horizon Chips JU EU Funded Project): Heterogeneous Integration for Connectivity and Sustainability – Call 2021 KDT JU – HORIZON-KDT-JU-2021-1-IA

The HiCONNECTS project develops the next generation of electronic components and systems heterogeneous integration core technology solutions for energy-efficient, high-performance wireless/wired cloud and edge computing and automotive radar in support of the next radio frequency and IT technology evolution. The high-level objective of HiCONNECTS is to support major societal and industrial challenges by developing heterogenous integration (HI) core technology solutions for energy-efficient and highperformance wireless/wired cloud and edge computing as well as automotive radar to solve two of the main state-of- the art roadblocks: the transmission of Internet of Things (IoT) data over the IT network and sensing of objects to enable Highly Automated Driving (HAD). Today, high latency, limited bandwidth, and high energy consumption per bit of information of current wireless/cloud platforms cannot satisfy the IoT applications that require real-time response. Furthermore, a lack of resolution, classification, segmentation, localization, and mapping of current radars is not sufficient for Level 4 and 5 HAD. The second and directly linked objective is to use these HI components to develop solutions in societal and economical important areas, with major focus on RF infrastructure (networking cards and switches) for the advancement of vertical sectors such as connected technologies (5G/6G and radar) and IT data growth vehicles, smart cities, and connected industry.

ADAS+ - Advanced Driver Assistance System

Recently, the ability to monitor the level of attention of a car-driver has attracted a great deal of interest, in the context of the automotive industry, due to the possibility to prevent any dangers that could arise from an inadequate psycho-physical state of the car driver. In terms of development, that recalled issue required Computer Vision development to constitute an effective ad-hoc real time warning system. Firstly, the researchers focused their effort on the development of integrated devices capable of evaluating not only the road route but also the position of the vehicle, the speed, the sudden change in acceleration and so on. Lately, the need to implement increasingly sophisticated machine learning methods has given more credit to the work based on the study of the psycho-physical condition of the driver as a considerable support for the detection of drowsiness. The intuition that underlies these methods is to verify the attention threshold through the so called heart rate variability (HRV) i.e. a complex physiological indicator for drowsiness. The analysis of HRV allows to understand the state of activity of the autonomic nervous system which is designed to manage a series of automatic, unconscious and involuntary activities, relating, for example, to heartbeat, blood pressure etc ... This variability can be calculated with the aid of appropriate physiological signals: the electrocardiographic (ECG) signal and the photoplethysmographic (PPG) signal. The proposed method has the purpose to show the existence of such correlation between the PPG signal, obtained through appropriate sensors, and the PPG signal reconstructed using the landmarks extracted from a face image. For the recording of biometric signals such as PPG, it will no longer be necessary to make use of integrated devices (sensors, etc.) but it is needed a simple low frame-rate video camera, resulting in a consequent saving in terms of both costs and resources.. For this purpose, we studied the skin micromovements and changes in facial color due to blood circulation quite indistinguishable with naked eye and related to the circulation controlled through the activity of sympathetic and parasympathetic nerves strongly regulated by autonomic nervous system.

(2025)

WORKSHOP@ICCV 2025: PHAROS - Adaptation, Fairness, Explainability in AI Medical Imaging Workshop (PHAROS-AFE-AIMI) and Competition to be held in conjunction with the International Conference on Computer Vision (ICCV), 2025 in Hawaii, USA, 19 – 23 October, 2025.

A main target of the Workshop is to present and evaluate novel approaches for predictive modeling of big medical image datasets, focusing on deep learning models, on transparent and human-centered integration of Generative AI and Large Models (such as LLMs, VLMs) in Health Services, creating trust among citizens. It addresses key challenges at the intersection of computer vision and healthcare AI, including multidisease diagnosis, model explainability, fairness, domain adaptation, and continual learning. Given the increasing interest in trustworthy AI, interpretable deep learning models, and the responsible deployment of AI in sensitive applications, this workshop will foster discussions on critical advancements shaping the future of AI in medical imaging. The Workshop is organised in compliance with the PHAROS AI Factory, which is one of the first seven AI Factories selected and funded by the European Union. This ensures that the workshop’s topics have real-world applicability and a strong foundation in cutting-edge research. More information can be found on the workshop's website.

AI Medical Image Analysis - ICCV 2025 Workshop - PHAROS

Coming soon - Work in Progress

AI4Industry Legal and Financial (AI4ILF) Research Group @ IPLAB