The Data & AI Systems (DAI) group spearheads the development of interdisciplinary methodologies pertinent to the design and implementation of intelligent systems, which utilize Artificial Intelligence, Machine Learning, and transformative data technologies with an emphasis on tangible impact.
The vision of DAI is to reconcile the specificities of data and AI with the practical demands of industries and societies, facilitating the provision of accessible and effective solutions that foster advancement and innovation. In pursuit of this objective, the DAI team concentrates on overcoming the existing constraints of data and AI, expanding the boundaries of potentiality, and devising next-generation solutions that will shape tomorrow's world.
The Data & AI Systems (DAI) group at UBITECH focuses its expertise on the convergence of big data technologies and artificial intelligence, thereby developing intelligent systems capable of learning, adapting, and facilitating informed decision-making across various industrial sectors. Our mission is to drive the advancement of AI applications in a responsible and ethical manner, ensuring that developed solutions are both potent and in harmony with societal values. The group is engaged in: (i) designing intelligent systems aimed at optimizing resource allocation and enhancing efficiency within multiple industries while delivering AI-powered solutions focused on sustainable development predicated on privacy-preserving AI technologies; (ii) the development of ethical and transparent AI models suitable for civilian applications and sensitive data environments; (iii) investigating and applying explainable AI (XAI) methodologies to promote transparency and trustworthiness in AI-based decision-making processes.
Key Research Areas:
- Privacy-Preserving Artificial Intelligence for Sustainable Development: We are pioneering the development of AI algorithms and systems that can analyze and learn from large datasets while ensuring the privacy and security of sensitive information. This research is directly applied to creating intelligent solutions for sustainable resource management, environmental monitoring, and optimized energy consumption across various industries in Greece and beyond.
- Ethical and Transparent AI for Civilian Applications: Our research focuses on the creation of AI models that are not only highly performant but also inherently ethical and transparent. This involves investigating and implementing fairness metrics, bias detection and mitigation techniques, and the development of AI systems that are auditable and trustworthy for use in civilian applications, including those handling sensitive data within the European regulatory framework.
- Explainable Artificial Intelligence (XAI) for Trustworthy Decision-Making: A core research area involves the investigation and application of Explainable AI (XAI) methodologies. We are developing techniques that allow us to understand and interpret the decision-making processes of complex AI models. This research is crucial for building trust in AI-driven insights, particularly in critical applications where accountability and transparency are paramount for adoption across Greek industries and public services.
Technological Domains
The Data & AI Systems (DAI) group develops ethical, privacy-preserving, and trustworthy AI systems, strategically leveraging a dynamic landscape of modern AI applications and cutting-edge technologies. Many of the current and future challenges in AI necessitate a deep understanding of these foundational concepts. Issues like data scarcity in privacy-preserving scenarios, the "black box" nature of deep learning models, and the need for efficient learning in complex environments require innovation at the algorithmic and problem-solving level. Our key technological domains encompass:
- Foundational Artificial Intelligence & Machine Learning: Core technological areas including supervised, unsupervised, and reinforcement learning, deep learning architectures (e.g., Convolutional Neural Networks, Recurrent Neural Networks, Transformers), and statistical modeling are being used. We explore both established techniques and novel approaches to build robust and adaptable AI models. Technologies used: TensorFlow, Keras, and PyTorch.
- Responsible Data Engineering & Management for Green AI: We focus on scalable data pipelines, data integration and fusion, feature engineering, data quality assessment, and the development of efficient data storage and retrieval solutions optimized for AI workloads. This includes expertise in big data technologies, modern data lake architectures and extensive benchmarking over big data pipelines to correlate computing and energy allocation with performance metrics. Technologies used: Apache Spark, Apache Kafka, Trino, MySQL/PostgreSQL, MongoDB, Minio, Neo4J.
- Privacy-Enhancing Technologies (PETs): A core commitment to ethical AI drives our research into PETs, including federated learning, differential privacy, and homomorphic encryption for queryable data lakes and secure learning on encrypted data. We have developed a Privacy-preserving Query Engine (PQE) and a Private AI/ML Operations Flow Engine (AIOpsFlow) that respect user privacy while delivering valuable insights. Technologies used: Trino, Zama, Concrete-ML.
- Natural Language Understanding (NLU) & Conversational AI: We leverage the power of language through research in areas such as natural language understanding, natural language generation, text summarization, sentiment analysis, topic modeling, and machine translation to contextualize Knowledge Graphs from unstructured text, generate networks with actors and activities in space, time and correlated with thematic areas. The Knowledge Graphs augmented with semantic technologies allows for knowledge representation, and reasoning. Technologies used: Transformers (Hugging Face), SpaCy, Neo4J, NetworkX.
- Explainable AI (XAI) & Trustworthy AI: We develop methods for making AI decision-making processes transparent and interpretable. This includes techniques for feature importance analysis, counterfactual explanations, saliency maps, and the development of metrics for fairness, explainability with scores, robustness, and transparency. Technologies used: LIME, SHAP, Grad-CAM.
- Edge AI & Cloud Integration: We investigate the deployment and optimization of AI models on edge devices and their seamless integration with cloud ecosystems, addressing challenges related to resource constraints, quality, and real-time processing. Our approach enables more efficient and lightweight models for edge deployment, and the exploration of novel architectures for specific tasks (e.g., tiny models for edge processing, few-shot learning, graph neural networks for relational data). Technologies used: LIME, SHAP, Grad-CAM, Adversarial Robustness Toolbox (ART).
Specialized Expertise
The Data & AI Systems (DAI) group specializes in a unique blend of deep theoretical understanding and practical, ethically-driven application of AI in real-world systems. Our core competencies extend beyond the mere utilization of specific technologies and are rooted in our methodologies, problem-solving approaches, and areas of deep knowledge:
- Holistic and Ethical AI Development Lifecycle: The team possesses a comprehensive understanding of the entire AI development lifecycle, from responsible data sourcing and engineering ("Green AI") to ethical model building (privacy-preserving techniques) and ensuring trustworthiness (explainability and robustness) post-deployment (Edge AI with cloud integration). This holistic view allows us to address challenges systemically rather than isolation. Projects: CYGNUS, TALON, RAIDO, CEDAR, PROTEAS
- Deep Algorithmic Understanding: Our strong grounding in foundational AI and ML (spanning supervised, unsupervised, reinforcement learning, deep learning, and statistical modeling) allows us to not only leverage existing algorithms but also to adapt and innovate at the algorithmic level to tackle specific challenges like data scarcity and the "black box" problem. Projects: TRITON, AppTake
- Problem-Driven Innovation: We are adept at identifying real-world problems, particularly those hindering the adoption of digital technologies (as highlighted in the initial context), and tailoring our AI expertise to develop practical and effective solutions. The focus on automating the software development lifecycle is a prime example of this problem-solving approach. Projects: Ceasefire, MobiSpaces, AVALANCHE, PRESERVE, AURORA
- Pragmatic Edge AI Deployment with Cloud Synergy: We have deep knowledge in optimizing AI models for deployment on resource-constrained edge devices, including resource-aware model optimizations and architectures for efficient and lightweight deployments. Projects: CYGNUS, TALON, RAIDO, PROTEAS
Description
APPtake aims to secure the operation and business sustainability of European SMEs, building a portfolio of sustainable application security solutions to foster the adoption of DevSecOps practices by the European SMEs, increasing their cybersecurity capabilities. APPtake will make available a marketplace platform to support interaction between suppliers and adopters of cybersecurity solutions and where the solutions of the seven technology providers in the project will be made available. The project will include both supply and demand support measures, implementing a maturity and integration phase of the solutions (supply) and featuring a massive demonstration plan with six end-users each from a different sector plus a hybrid online plus in-presence strategy to reach commercial partners and customers (demand). The project takes extensive advantage of the outcomes of past publicly-funded research in the EU, both for the marketplace and for the solutions, several of which have developed in the framework of EU-supported research and innovation projects.Key Contributions
UBITECH makes available to the APPtake ecosystem the DevSecOps Marketplace which will author the entire lifecycle of deployment and runtime of: (i) identity and access management services; (ii) data loss prevention services; (iii) web and apps security services; (iv) email security; (v) cybersecurity risk assessment; (vi) intrusion management; (vii) SIEM; (viii) encryption; (ix) business continuity disaster recovery as a service; and (x) network security. UBITECH will extend and capitalize over its past experience w.r.t. cybersecurity marketplaces bringing its maturity from APPtake (https://puzzle-h2020.com) and SecureIoT (https://secureiot.eu/) projects. The challenges and limitations that will be addressed in APPtake lie within the realization and grounding of such solutions with real-world cases. The APPtake Marketplace will support (i) multi-roles; (ii) registering participants and business entities; (iii) search, access and discovery of DevSecOps service offerings; and (iv) a rich catalogue of DevSecOps services with management, auditing and tracking functionalities.
Description
AURORA aims to enhance resilience against health risks stemming from climate change, by following a six-step approach, developing tools capable of 1) monitoring climate stressors, 2) developing climate and epidemiological models, 3) generating forecasts through simulated scenarios, and 4) identifying risks and vulnerabilities related to a) anticipated changes in the climate and geography of Baltic countries, and b) health, well-being and socio-economic stressors affected by climate change. Finally, the assessed risks will be accompanied by trustworthy artificial intelligence (AI)-driven technology to 5) generate early warnings and 6) recommend adaptations and nature-based solutions, forming a robust Decision Support system for health impacts of climate change. AURORA’s generated outputs will ultimately support Policy Makers, Healthcare professionals and Local Authorities to: a) enhance the resilience of their cities by designing appropriate adaptations, b) safeguard the one health system of EU’s Boreal region, and c) form efficient policies/strategies. The effectiveness of the AURORA toolset will be demonstrated and validated in five major Baltic cities (Riga, Vilnius, Tallinn, Tampere, Pori) and its results will be replicated in three participating municipalities (Klaipeda, Jurmala, Joniskis).Key Contributions
In AURORA, UBITECH implements the Simulation Engine, a powerful digital tool capable of executing the simulation of the scenarios generated by the scenario builder based on the climate, epidemiological and socio-economic models developed by the experts, the models of the relevant adaptations, as well as the vulnerabilities and risks of the area/city. The Simulation Engine harnesses the power of machine learning and artificial intelligence technologies to optimise its analytical capabilities. These technologies enable the algorithms and statistical models within the Engine to perform effectively, driving the generation of insightful and complex epidemiological scenarios driven by climate stressors. AURORA's Simulation Engine empowers users to simulate a wide range of epidemiological scenarios by configuring specific parameters. These parameters include phenomena, epidemiological impacts, risk thresholds, and temporal horizons. This user-friendly customisation ensures that the scenarios align with the unique characteristics of the demonstration city/region. The Simulation Engine will generate predictive and simulation models that run seamlessly in the background. These models simulate various scenarios based on climate stressors and their interactions with epidemiological factors. The scenarios are categorised by their likelihood, severity, and favourability, providing users with valuable insights into the potential outcomes and risks. AURORA's Simulation Engine will enhance the decision-making process by providing stakeholders, including epidemiologists, policymakers, and health professionals, with data-driven insights. It enables the assessment of potential health hazards resulting from climate change, informs adaptation strategies, and facilitates proactive risk mitigation. Beyond decision support, the Simulation Engine also functions as a valuable training tool. It employs data from other modules within the AURORA platform, along with aggregated data and metadata, to train embedded machine learning models. This prepares personnel for real-world scenarios and ensures preparedness for complex health challenges driven by climate stressors.
Description
AVALANCHE aims to make transformative steps towards the development of a highly innovative, holistic, multi-disciplinary, high-tech, and versatile solution for significantly increasing/broadening the operational capabilities of Law Enforcement Agencies (LEAs) in their struggle to detect, analyse, track, investigate and prevent cross-border illicit activities of high-risk criminal networks, migrant smuggling, firearms, drugs, child exploitation, terrorism, cyber and intellectual property crime coordinated through the digital world. The AVALANCHE platform will offer tools for semi-automatic collection of evidence to foster explainable investigations and reasoning through intelligence-led methods and contextual-aware gleaning of actions. Also, the AVALANCHE platform will provide interoperable systems and interconnection with national and international databases through common standards, data and schemas alignment, and mediation to foster information exchange and international cooperation. To guarantee the wide SMEs solutions adoption by the actual practitioners, AVALANCHE will produce training curricula, benchmark with Europol’s Innovation Lab, contribute to LEAs operational standards and establish a broad ecosystem of crime-relevant stakeholders, and complement its objectives by continuously receiving guidance from the EU Policy Instrument (EMPACT).Key Contributions
Apart from the overall project management and consortium coordination of AVALANCHE, UBITECH is responsible for the design of the software/hardware architecture of AVALANCHE (e.g., interfaces, information flows, component interactions, deployment views, etc.), following an Ethical-by-Design, Human-centred and Co-development approach. It will specify the logical structure of the AVALANCHE platform, emphasizing the communication between different solutions and system components. The solutions and system components will be selected and individually specified in a comprehensive manner, while implementation/integration will proceed based on an integrated view of the envisioned platform targeting the HP infrastructure. Moreove, UBITECH will implement develop a multi-threaded web crawling and intelligence system with a high degree of configurability and semantic awareness. Its goal will be to collect, harmonise and generate analytics for activities related to: (i) illegal goods trafficking (e.g., firearms, drugs, etc.); (ii) human and child exploitation (e.g., forums, dialogues, encrypted coordination channels, pornographic material, etc.); and (iii) illegal cyber activities (e.g., stolen documents, breached data, hate crimes or cyber-terrorism events, antisemitic rhetoric, racism, revenge porn, stolen credentials, orchestrated attempts and massive calls for DDoS attacks) from the surface and dark web, darknet marketplaces and social media. The data collection, storage (for data backup) and investigative analytics will be based on a human-in-the-loop approach, involving automated form-filling techniques, crawling configurations initiated by the end users and manual ad hoc queries constructed incrementally using relevant keywords, spatial or temporal criteria. Moreover, UBITECH will develop and deploy sophisticated data analysis pipelines in the following manner: (i) Advanced analytics will be applied to the constructed supergraph data structures for extracting node/entity clusters/groups, latent data properties, complex patterns, etc.; (ii) causal modelling algorithms (e.g. probabilistic graphical networks) will estimate/capture the hierarchical interrelations of the data; and (iii) Unsupervised learning (e.g. deep neural network autoencoders, etc.) will be applied to identify complex patterns and hidden correlations; so as to detect, extract, and model the complex (inter) relations among the set of available data points, targeting to identify the correlations between the recognized physical entities (i.e., suspect identities, IP addresses, encoded names, and locations) and eventually reconstructing (and monitoring) the acting criminal networks. The latter will include the analysis along the full chain of trade, i.e., focusing on detecting the criminal actors involved in the trafficking of illegal goods and the exploitation of human beings and the eventual commercial transactions.
Description
Ceasefire targets the development of a highly innovative, holistic, multi-disciplinary, high-tech and versatile approach for significantly increasing/broadening the operational capabilities of EU Law Enforcement Agencies (LEAs) in their struggle to detect, analyze and track cross-border illicit firearms trafficking related activities. In particular, Ceasefire efforts will concentrate on the following major driving axes: a) The development of advanced Artificial Intelligence (ΑΙ) technologies for significantly facilitating the everyday work of the involved practitioners, and b) The establishment of fully-operational National Focal Points (NFPs), by targeting to alleviate from current organizational, operational, cooperation, legal, cross-jurisdictional, trans-border and information exchange challenges. The former corresponds to a wide set of AI-enabled tools, including solutions for cyber patrolling, Web data gathering, on-the-spot detection of firearms, advanced Big Data analytics, cryptocurrency analysis, large-scale information fusion, visual analytics and firearms-related intelligence collection.Key Contributions
In Ceasefire, UBITECH undertakes the development of the open source intelligence collection and real-time systematic firearms incident data gathering tool, constituting a highly configurable, semantically-aware and multi-threaded web crawling system for retrieving data regarding firearm incidents (e.g. seizures, shootings, deaths, suicides, homicides, etc.) from open sources (i.e. news articles, press releases, blogs, forums, social media, online repositories, etc.). Moreover, UBITECH drives the implementation of the dark web and darknet marketplace analysis tools, that realize a highly configurable, semantically-aware and multi-threaded dark web crawling system designed to navigate (via web personas) in dark Web forums and darknet marketplaces.
Description
CEDAR will (1) identify, collect, fuse, harmonise, and protect complex data sources to generate and share 10+ high-quality, high-value datasets relevant for a more transparent and accountable public governance in Europe; (2) develop interoperable and secure connectors and APIs to utilise and enrich 6+ Common European Data Spaces (CEDS); (3) develop innovative and scalable technologies for effective big data management and Machine Learning (ML) operations; (4) deliver robust big data analytics and ML to facilitate human-centric and evidence-based decision-making in public administration; (4) validate the new datasets and technologies (TRL5) in the context of fighting corruption, thus aligning with the EU strategic priorities: digitalisation, economy, democracy; (5) actively promote results across Europe to ensure their adoption and longevity, and to generate positive, direct, tangible, and immediate impacts.Key Contributions
Within CEDAR, UBITECH coordinates the identification and definition of the open and proprietary data sources, repositories and collections to be used in the project, which will further guide the design of CEDAR data flows within the conceptual architecture and thedevelopment of CEDAR Data- and ML- Ops, as well as the data to be shared / integrated with existing CEDS. Moreover, UBITECH will significantly contribute to the Data Modelling, Harmonisation and Alignment mechanisms that will leverage semantic technologies to define the data models providing a common data structure to be used by different use cases, as well as to the Data Protection and (Pseud)Anonymisation mechanisms that will protect the personal data identified incorporating several data pseud-anonymization techniques including masking, generalisation, and perturbation, and assessment methods with regards to the re-identification risk and how to mitigate it.
Description
CYGNUS proposes a self-managing software platform putting the human-in-the-loop in the decision-making process, which provides sovereign capability for civilian applications and exploits the full potential of Causal AI, IoT / Edge Computing, Distributed and Big Data Processing technologies to establish accelerated AI automation with advanced autonomy, learning and agency characteristics. The CYGNUS Platform will implement and support a bouquet of big data and AI services integrated in a self-managing software platform putting the human-inthe-loop in order to: (i) sense, harvest, harmonise and aggregate huge amounts of sensorial data from distributed IoT devices and UAVs; (ii) train federated and centralized AImodels capable of automatically detecting and identifying the presence of significant and interesting entities, and behavioural patterns including normal and abnormal classification with spatiotemporal and contextual criteria and their correlations; (iii) reduce the analysis load performed by human operators via scalable big data processing modalities and Explainable AI(XAI) for justified decision making; and (iv) deliver a Multi-Objectives (M/O) Optimisation Solver for causal AI to derive cause-and-effect relationships, go beyond correlation-based predictive models and toward AI systems that can prescribe actions more effectively and act more autonomously.Key Contributions
Within CYGNUS, UBITECH undertakes the overall administrative, scientific and technical coordination, and drives the implementation of the Big Data & Accelerated AI Automation infrastructure. In particular, UBITECH designs and implements the big data operations and reusable containerized microservices (aka Big Data Workbench) for data acquisition, aggregation, harmonisation, processing and querying and is responsible to set up the message brokers (i.e., Kafka, RabbitMQ) and the data bridges with the CYGNUS in-memory at edge and in-disk at cloud storage system, supporting also data cleaning, harmonisation and quality improvement by modelling and harmonising in a unified model the raw and aggregated data from UAVs, videos, and sensors. Moreover, UBITECH designs and implements the XAI methods extended with a Probability Score (+2 Likert scale) in order to enable Causal AI with justified predictions on objects, patterns, and events / actions / incidents, eliminate false positives and let human operators interpret the decision paths that the results of AI models are producing, by extending the open-source libraries SHAP and LIME. Finally, UBITECH designs and implements the adversarial robustness methods (i.e., by extending the open-source library ART) of CYGNUS via adversarial ML training to tackle models' evasion, while novel computer vision transformers and tensor-based learning will increase AI model response robustness to statistical fluctuations of input data.
Description
MobiSpaces aims to deliver an end-to-end mobility-aware and mobility-optimized data governance platform with key differentiating factor that the outcomes of mobility analytics will be utilized to optimize the complete data path, in terms of efficient, reliable, secure, fair and trustworthy data processing. MobiSpaces promises the extraction of actionable insights from ubiquitous mobile sensor data and IoT devices in a decentralized way, offering intelligent transportation services, enforcing privacy constraints at the expected point of action. XAI techniques will be applied at the level of data management and machine learning, supporting the creation of comprehensive and interpretable prediction models, while all the research outcomes will be validated through five use cases: 1) intelligent public transportation services in urban environments, 2) intelligent infrastructure traffic sensing for smart cities, 3) vessel tracking for non-cooperative vessels, 4) decentralized processing on[1]board of vessels, and 5) enhanced nautical maps via crowdsourced bathymetry vessels data. These actions will be continuously adapted and monitored for their environmental sustainability, whereas MobiSpaces will create a widely accepted standard of data processing and analytics, alongside a data-rich ecosystem providing trustworthy and actionable data that is vital for enabling the growth of the EU digital economy.Key Contributions
Within MobiSpaces, the contribution of UBITECH focuses on the data management aspects, related with: (i) the delivery of mobility-aware AI and pattern analysis algorithms, (ii) the provision of adaptive resource allocation and configuration, (iii) the performance of efficient and user-friendly declarative analytics, (iv) the delivery of AI-enhanced aggregation algorithms for resource-constrained edge nodes, (v) the implementation of an overall decentralized data management system, for supporting decentralized data operations, and (vi) the protection of data flows.
Description
In the era of digital transformation, criminal activity has evolved dramatically, presenting complex challenges to law enforcement authorities (LEAs). The proliferation of social networks, the expansiveness of the deep and shallow web, and the abundance of data available online have become tools for illicit and harmful activities, and criminals can now make use of digital tools to achieve their goals. The EU-funded PRESERVE project will develop an innovative, privacy-preserving decision-support system for European LEAs using AI and big data to combat cybercrime and terrorism. The project will use a range of cutting-edge AI methodologies and techniques to monitor social networks, police databases, and web information, identifying communities and users involved in hate speech, child sexual abuse, terrorism, and drug trafficking activities.Key Contributions
Within PRESERVE, UBITECH drives the development of the PRESERVE data crawler, a highly configurable, semantically aware, and multi-threaded web crawling and intelligence system designed to monitor and analyse illegal activities across social media, surface, deep, and dark web. The system will focus on collecting, harmonising, indexing, and serving electronically sealed data for analysis, which can be used in court. The system employs a user-assisted approach that includes automated form-filling and manual queries using keywords from the project's ontology, and spatial or temporal criteria. A distinctive feature is its genetic algorithm, which refines and optimises search keywords and content, targeting more relevant URLs, particularly on dark websites.
Description
PROTEAS aims at providing a prototyped deployable Special Operations Forces (SOF) Command Post (SOFCP) Command & Control (SOFCPC2) consisting of the following parts: (a) the SOFCP shelter that will be air/sea/ground transportable, rapidly deployable and resilient to adverse operational and weather conditions; (b) the SOFCP Power system that will be a military grade, autonomous, horizontally scalable and low thermal/acoustic signature system capable of furnishing the energy needs of the entire SOFCP; (c) the SOFCP interoperable and cyber-secure communications infrastructure that will enable the exchange of information across the entire command hierarchy; (d) the multiple domain Command and Control (C2) capable of supporting operations related to intelligence gathering from various sources, analysis and further dissemination of them towards enhanced situation awareness; (e) the portable devices for field deployed SOF task groups providing C2 and DACAS (Digitally Aided Close Air Support) functionalities at the tactical edge; (f) the manned and unmanned assets, employed either for transportation of SOF Task Groups or the collection of intelligence and (g) the SOFCP perimeter security system capable of detecting ground threats near the camp.Key Contributions
Within PROTEAS, UBITECH contributes towards the design and development of PROTEAS Intelligence & Sensing Software Platform for real-time processing and correlation (based on the injected information from all sensors deployed by the SOF Task Units operating on the field). In particular, UBITECH will collect datasets (real and/or synthetic ones) to train AI algorithms; design a Big Data Analytics Framework for big data fusion, processing and storage of data provided from heterogeneous sources; design AI algorithms to support SOF Small Joint Operations (SJO) missions and specifically SOF Units located in All-Source Center within SOFCPC2; design Human Machine Interfaces for SOF Team Analysts to access data from Intelligence Engine; design an AI- based Federated network infrastructure to support data exchange of different classification levels.
Description
RAIDO is a powerful framework solution designed to develop trustworthy and green artificial intelligence (AI). Trustworthy AI focuses on ensuring the reliability, safety, and unbiased optimization and deployment of AI systems, particularly in critical applications such as healthcare, farming, energy, and robotics. On the other hand, Green AI involves the development and deployment of energy-efficient and environmentally sustainable AI technologies, leading to reduced environmental impact and improved resource management. RAIDO provides an array of automated data curation and enrichment methods, including digital twins and diffusion models, to create high quality, representative, unbiased, and compliant training data. It also offers various data- and compute-efficient models and tools to create energy-efficient Green AI, such as few- and zero-shot learning, dataset and model search, data and model distillation, and continual learning. To ensure the transparency, explainability, and reliability of the optimized AI models and data handling processes, RAIDO uses various XAI methods, decentralized blockchain, feedback-based reinforcement learning, novel KPIs, and visualization techniques. Additionally, the innovative AI orchestrator optimizes related tasks and processes, reducing the overall energy consumption and environmental footprint of the models during both development and deployment. RAIDO emphasizes the development of dynamic interfaces that support the appropriate AI paradigms (central, distributed, dynamic, hybrid) and enable seamless adaptation to the needs of the use situation. Furthermore, RAIDO will be evaluated through four real-life demonstrators in key application domains, such as smart grids, computer vision-based smart farming, healthcare, and robotics, showcasing notable societal and market impact.Key Contributions
Within RAIDO, UBITECH will drive the definition of RAIDO's innovative architecture, resulting in reliable AI and data optimisation systems, as well as the creation and development of RAIDO's novel data generation and AI optimisation framework, which is built on data curation, lifelong learning, transfer, and few-shot learning, and feedback and monitoring processes. It supports data distillation, HITL, XAI, secure data storage, and ethical AI to describe and evaluate how AI algorithms work in an effort to lessen the need for large volumes of data by providing more energy- and data-efficient AI models. Moreover, UBITECH will implement RAIDO’s Data Enrichment, Curation and Distillation, and Federated Mining pipeline building all the essential components in order to improve the quality and the quantity of the data, incorporating filtering techniques (e.g., handle of missing data, outlier detection, and noise removal); harvesting and feature extraction techniques, (e.g., feature engineering, dimensionality reduction); balancing techniques (e.g., oversampling, decision threshold adjustment, class weight adjustment); transformation, smoothing, normalization, and attribute construction techniques; data annotation tools for (a) image data (e.g. keypoint tool, auto-annotation tool, polygon tool for semantic segmentation); and (b) text and timeseries, (e.g. semantic annotation, intent annotation, and the addition of metadata). Finally, UBITECH will deliver RAIDO's Information Layer by implementing Data Lakes for storage and efficient AI training with main goal to increase the efficiency both in the training and the deployment of AI models.
Description
TALON aims at sculpturing the road towards the next industrial revolution by developing a fully-automated AI architecture capable of bringing intelligence near the edge in a flexible, adaptable, explainable, energy and data efficient manner. TALON architecture consists of three fundamental pillars: a) an AI orchestrator that coordinates the network and service orchestrators in order to optimise the edge vs cloud relationship, while boosting reusability of datasets, algorithms and models by deciding where each one should be placed; b) a lightweight hierarchical blockchain schemes that introduce new service models and applications under a privacy and security umbrella; and c) new digital-twin empowered transfer learning and visualization approaches that enhance AI trustworthiness and transparency. It combines the benefits of AI, edge and cloud networking, as well as blockchain and DTs, optimized by means of a) new key performance indicators that translate the AI benefits into insightful metrics; b) novel theoretical framework for the characterisation of the AI; c) blockchain used to deliver personalised & perpetual protection based on security, privacy and trust mechanisms; d) AI approaches for automatically and co-optimising edge and cloud resources as well as the AI execution nodes; e) semantic AI to reduce the learning latency and enhance reusability; and f) digital twins that visualize the AI outputs and together with human-in-the-loop approaches.Key Contributions
Within TALON, UBITECH undertakes the technical, scientific and innovation management of the TALON action, and drives the activities towards the design and development of TALON’s novel theoretical framework that employs information, probability and game theory approaches to explain and asses the operation of AI algorithms. Last but not least, UBITECH will define and design the NG-SDN and distributed intelligence functionalities of TALON framework. The NG-SDN toolkit offers top-down programmability and embraces white-box devices, enabling the operator to interact with the network in a manner previously impossible by defining custom solutions and protocols, manipulating packets in a new fashion. Through technologies like P4 and XDP, the operator can program the desired behaviour in multi-vendor devices with little effort, achieving end-to-end control of the network.
Description
TRITON‘s vision is to overcome defence-specific obstacles associated to the automation of penetration tests, and fully automa te the pre- and post- pentesting process adopting Markov chain Monte Carlo (MCMC) decision processes to discover hidden attack paths and the DevSecOps paradigm integrated with well-known pentesting frameworks (e.g., Kali Linux, Aircrack-ng, Metasploit, etc.) focusing on military Security Operation Centres (SOCs), web and heterogeneous cloud applications, telecom and wireless networks. It introduces the novel concept of Human-as-a-Security-Sensor (HaaSS), letting the operators of the automated penetration testing solution to monitor the progress, perform what-if analysis, predict future paths, and enforce controls through security policies. The key idea is to build a wide‐ranging manifold of novel tools and strategies that enable next‐generation ICT systems and networks with distributed devices, to perform automated and Artificial Intelligence (AI) driven security assessments at massive scale. TRITON is expected to realize its vision through generative AI and ethical attacks performed by Generative Adversarial Networks (GANs) and proactive risk assessments with optimal mitigation controls targeting code, firmware, networks, and ICT deployment environments.Key Contributions
UBITECH leads Task 2.2 (T2.2) on “Automated Penetration Testing Methodology in Military SOCs, Networks, Web, and Systems.” This task focuses on adapting established security assessment methodologies, such as ENISA skillset recommendations and NIST’s Information Security Testing and Assessment frameworks. TRITON will extend these methodologies by introducing novel, automated steps for penetration testing in military contexts and isolated execution environments. The new TRITON Methodology will incorporate best practices from DevSecOps, generative AI attack strategies, and intelligent game-theoretic optimizations, enhancing the effectiveness of security assessments in defense operations. In WP4, UBITECH leads Task 4.1 on “Knowledge Base & Persistence, Data Collection Tools, Graph-based Attack Scenarios & Reporting.” This task aims to develop a sophisticated network Attack Configuration and Emulation Engine (ACEE) that combines AI/ML techniques and emulation capabilities to simulate attacker and defender actions. By leveraging ARM’s Pentest Suite and UBITECH’s OLISTIC cybersecurity risk management suite, the ACEE will be able to identify attack paths and train Deep Reinforcement Learning (DRL) agents to discover hidden vulnerabilities. UBITECH also leads Task 4.2 on “Model-based Weaknesses and System-based Vulnerabilities Identification.” This task integrates the TRITON Model with renowned cybersecurity databases such as MITRE CPEs, CVEs, CWEs, and NVD to identify potential threats to target systems. The task will utilize the Product and Firmware Pentesting (PFP) engine, enabling massive automated penetration testing on active products and assets. The PFP engine will employ a model-based approach, utilizing various pentesting modules to conduct vulnerability assessments across networks, systems, web apps, containers, and cloud infrastructures. As part of Task 6.4, UBITECH will also focus on “Post-exploitation, Privilege Escalation, and Lateral Movement.” This task will employ advanced tools such as Meterpreter (from Metasploit), Empire, and CrackMapExec (CME) to simulate post-exploitation techniques, including privilege escalation, lateral movement, and domain infiltration. These tools will enhance the ability to assess and mitigate security threats in a real-world, dynamic defense environment.
Description
The BigDataStack H2020-779747 project aims to deliver a complete high-performant stack of technologies addressing the emerging needs of data-intensive operations and applications, which is based on a frontrunner infrastructure management system that drives decisions according to data aspects thus being fully scalable, runtime adaptable and performant for big data operations and data-intensive applications. BigDataStack promotes automation and quality and ensures that the provided data are meaningful, of value and fit-for-purpose through its Data as a Service offering that addresses the complete data path with approaches for data cleaning, modelling, semantic interoperability, and distributed storage. BigDataStack introduces a pioneering technique for seamless analytics which analyses data in a holistic fashion across multiple data stores and locations, handling analytics on both data in flight and at rest. Complemented with an innovative CEP running in federated environments for real-time cross-stream processing, predictive algorithms and process mining, BigDataStack offers a complete suite for big data analytics.Key Contributions
Within BigDataStack, UBITECH will drive the implementation of (a) the information-driven networking framework that will incorporate data-centric networking mechanisms combined with software defined networking technologies for their implementation and management over virtualised resources, delivering technologies for traffic engineering and network management for data operations and data-intensive applications; and (b) a toolkit allowing big data practitioners both to ingest their analytics tasks (through declarative methods for expressing the information extraction methods and predictive data services) and to set their requirements and preferences as quality metrics, affecting the application dimensioning, deployment patterns generation and prioritization, application components and data services configuration, quality monitoring and enforcement, and information-driven networking. Finally, UBITECH will contribute in the development of an homogeneous and coordinated data-driven orchestration of application components and data services, incorporating mechanisms to estimate and formally describe the synergies between the different infrastructure parameters, metrics and KPIs, and the application- and data- related aspects (e.g. used data analytics functions, data I/O volume, data flow predictions, etc.).
Description
The COALA H2020-957296 project aims at the innovative design and development of a voice-first Digital Intelligent Assistant for the manufacturing sector. The COALA solution will base on the privacy-focused open assistant Mycroft. It integrates prescriptive quality analytics, AI system to support on-the-job training of new workers, and a novel explanation engine - the WHY engine. COALA will address AI ethics during design, deployment, and use of the new solution. Critical components for the adoption of the solution are a new didactic concept to reach workers about opportunities, challenges, and risks in human-AI collaboration, and a concurrent change management process.Key Contributions
Within COALA, UBITECH undertakes the role of the integrator and develops the system architecture, data interfaces, APIs specification, and data security measures; manages the tools, methods, and integration issues through the definition of the integration points, the design of a detailed technical architecture, and the definition of the integration and testing plan; defines code maintenance and quality procedures to raise system reliability and exploitability; and creates the solution as an integrated whole with a security framework matching industry needs.
Description
The CYBELE H2020-825355 project aspires at demonstrating how the convergence of HPC, Big Data, Cloud Computing and the IoT can revolutionize farming, reduce scarcity and increase food supply, bringing social, economic, and environmental benefits. CYBELE intends to safeguard that stakeholders have integrated, unmediated access to a vast amount of large scale datasets of diverse types from a variety of sources, and they are capable of generating value and extracting insights, by providing secure and unmediated access to large-scale HPC infrastructures supporting data discovery, processing, combination and visualization services, solving challenges modelled as mathematical algorithms requiring high computing power.Key Contributions
Within CYBELE, UBITECH will undertake the responsibility of the technical coordination of CYBELE towards the realization of large scale HPC-enabled test beds and delivers a distributed big data management architecture and a data management strategy providing 1) integrated, unmediated access to large scale datasets of diverse types from a multitude of distributed data sources, 2) a data and service driven virtual HPC-enabled environment supporting the execution of multi-parametric agri-food related impact model experiments, optimizing the features of processing large scale datasets and 3) a bouquet of domain specific and generic services on top of the virtual research environment facilitating the elicitation of knowledge from big agri-food related data, addressing the issue of increasing responsiveness and empowering automation-assisted decision making, empowering the stakeholders to use resources in a more environmentally responsible manner, improve sourcing decisions, and implement circular-economy solutions in the food chain. Moreover, UBITECH R&D team will drive the implementation of a dedicated Experiment Composition Environment, enabling simulation execution in the Precision Agriculture and Precision Livestock Farming domains, that aims to facilitate the detaching of the design, development and execution of the big data analysis processes, supporting embedded scientific computing and reproducible research. The analysis process will be based on the selection of an analysis template, where each analysis template will represent a specific algorithm with the associated software and execution endpoint, and will provide to the user the flexibility to adjust the relevant configuration parameters, including input parameters for the algorithm, execution parameters, parameters associated with networking and computing resources constraints, as well as output parameters. The Experiment Composition Environment will support the design and implementation of data analysis workflows, consisted of a series of data analysis processes, interconnected among each other in terms of input/output data streams/objects. The analytics workflows designed, will be sent for execution on well-known HPC and Big Data frameworks, which will run on HPC resources abstracted to the user.
Description
The CyberSANE H2020-833683 project intends to improve the detection and analysis of cyber-attacks and threats on Critical Information Infrastructures (CIIs), increases the knowledge on the current cyber threat landscape and supports human operators (such as Incident Response professionals) to dynamically increase preparedness, improve cooperation amongst CIIs operators, and adopt appropriate steps to manage security risks, report and handle security incidents. Moreover, CyberSANE is fully in-line with relevant regulations (such as the GDPR and NIS directive), which requires organizations to increase their preparedness, improve their cooperation with each other, and adopt appropriate steps to manage security risks, report and handle security incidents. In particular, CyberSANE will develop a system that addresses both technical and congitive challenges related to identification, prevention and protection against attacks. At technical level, the CyberSANE system will collect, compile, process and fuse attack related data from multiple perspective, through its main four components: The Live Security Monitoring and Analysis (LiveNet) component, the Deep and Dark Web mining and Intelligence (DarkNet) component, the Data Fusion, Risk Evaluation and Event Management (HybridNet) component and the Intelligent and Information Sharing and Dissemination (ShareNet) component. From a cognitive perspective, the system will enable decision makers (e.g. incident response professionals) to better understand understand the technical aspects of an attack and draw conclusions on how to respond.Key Contributions
Within CyberSANE, UBITECH is one of the main technology providers to work towards the implementation of LiveNet that constitutes an advanced and scalable Live Security Monitoring and Analysis component capable of preventing and detecting threats and, in case of a declared attack, capable of mitigating the effects of an infection/intrusion. The main objective of this component is to implement the Identification, Extraction, Transformation, and Load process for collecting and preparing all the relevant information, serving as the interface between the underlying CIIs and the CyberSANE system. To this end, this component includes proper cyber security monitoring sensors including network-based Intrusion Detection Systems (IDS), innovative Anomaly detection modules and endpoint protection solutions for accessing and extracting information, on a real-time basis, in order to detect complex and large-scale attacks such as Advanced Persistent Threats (APTs). Moreover, UBITECH is one of the core partners developing the Deep and Dark Web mining and intelligence (DarkNet) component that provides the appropriate Social Information Mining capabilities that will allow the exploitation and analysis of security, risks and threats related information embedded in user-generated content (UGC). This will be achieved via the analysis of both the textual and meta-data content available from such streams. Textual information will be processed to extract data from otherwise disparate and distributed sources that may offer unique insights on possible cyber threats. Finally, UBITECH participates in the development of the Data Fusion, Risk Evaluation and Event Management (HybridNet) component that provides the intelligence needed to perform effective and efficient analysis of a security event based on: (i) information derived and acquired by the LiveNet and DarkNet components; and (ii) information and data produced and extracted from this component. In particular, HybridNet component retrieves incidents-related data via the LiveNet component from the underlying CIIs and data from unstructured and structured sources (e.g. from Deep and Dark Web) consolidated in a unified longitudinal view which are linked, analyzed and correlated in order to achieve semantic meaning and provide a more comprehensive and detailed view of the incident.
Description
The CYRENE H2020-952690 project aims to enhance the security, privacy, resilience, accountability and trustworthiness of Supply Chains (SCs) through the provision of a novel and dynamic Conformity Assessment Process (CAP) that evaluates the security and resilience of supply chain services, the interconnected IT infrastructures composing these services and the individual devices that support the operations of the SCs. In order to meet its objective, the proposed Conformity Assessment Process is based on a collaborative, multi-level evidence -driven, Risk and Privacy Assessment Approach that support, at different levels, the SCs security officers and operators to recognize, identify, model, and dynamically analyse advanced persistent threats and vulnerabilities as well as to handle daily cyber-security and privacy risks and data breaches.Key Contributions
Within CYRENE, UBITECH undertakes the overall technical management ensuring the correct performance of the project's technical tasks. Additionally, UBITECH drives the definition of the strategy of realizing the CYRENE Conformity Assessment proces as well as the CYRENE multi-level evidence-driven Supply Chain Risk Assessment process. In particular, UBITECH is responsible for the design and development of the three horizontal layers (HLs) dedicated to Risk and Privacy Assessment: “HL1-Risk and Privacy Assessment of Supply Chains”, “HL2-Risk and Privacy Assessment of ICT-based Supply Chains” & “HL3-Risk and Privacy Assessment of CIIs’ IT ecosystems including IoT ecosystems/devices / ICT Systems”. Finallly, UBITECH develops a highly configurable crawling service that facilitates crawl control, by either crawling all the data and information, process and post-process them to be ready for shallow and deep analytics or by performing “focused crawling” processes where given a concept, the services explore relevant parts of the Web and collects the required threats and risk-related data. Analytics libraries and algorithms (e.g. deep semantic analysis techniques together with NLP techniques) for knowledge extraction and business intelligence will be also used in order to extract meaningful knowledge that will be in a position to identify situations that can become a threat for the SCs under examination
Description
The LAW-GAME H2020-101021714 project aims to train police officers’ on the procedure, enhancing the transition between the theory and reallife practice through gamification technologies in a safe and controlled virtual environment. Essential tasks during the creation of LAW-GAME serious game are to virtualise and accurately recreate the real world, by realistically simulating and analysing aspects of a real-world situations. LAW-GAME introduces an attractive approach to the development of core competencies required for performing intelligence analysis, through a series of AI-assisted procedures for crime analysis and prediction of illegal acts, all within the LAW-GAME game realm. Building upon an in-depth analysis of police officers’ learning needs and inspired by a multitude of disciplines, LAW-GAME develops an advanced learning experience, embedded into 3 comprehensive “gaming modes” dedicated to train police officers and measure their proficiency in conducting forensic examination; effective questioning, threatening, cajoling, persuasion, or negotiation; and recognizing and mitigating potential terrorist attacks.Key Contributions
Within LAW-GAME, UBITECH’s participation has a two-fold focus: to create AI based algorithms to detect patterns indicating potential criminal and/or terrorist actions; and to create explainable (XAI) models, so as to ensure transparency, trustworthiness and explainability of the aforementioned illegal activities predictive models. In particular, UBITECH will work towards the AI-enabled criminal and terrorist attack modelling, though the implementation and validation of AI/ML based algorithms that will enable the Law Enforcement Agencies (LEAs) to understand and anticipate potential terroristic actions. Special attention will be given in training and deploying recurrent deep neural network algorithms (RNNs, Long Short Term Networks-LSTM) to learn the temporal context of the action-generated datasets (expressed as time-dependent sequences of recognised events) in order to make better predictions with respect to potential future terrorist actions. Specifically, the algorithms will provide recommendations to the officers about events and actions of the other players (e.g., terroristic group) that are potentially linked to terroristic activity, given past activities/ actions of the users during the course of the game. Last but not least, UBITECH will implement and validate a set of XAI techniques, which will ensure that the Machine Learning and Predictive Analytics techniques adopted for the illegal activities detection are either transparent or explainable so as to be trusted and acceptable by the users of the game.
Description
The MYRIOPOS T2EDK-04665 project aims to introduce innovative services, algorithms and solutions for: (i) the efficient scalability and management of large-scale distributed data, analysis and visualization, (ii) real-time network analysis and data traffic management through advanced deep learning and applied statistical algorithms for the timely detection of anomalies, behavioural outliers from the web and dark web based on network data and content coming from social networking media, forums and the internet; iii) the application of innovative solutions regarding the development of flexible and scalable approaches targeting at the distributed processing, fast networking mechanisms and data offloading to enable real-time analysis; and (iv) advanced data mining and intelligence mechanisms utilizing artificial intelligence and big data , in order to provide real-time detection, analysis and knowledge extraction specialized to security scientists, infrastructure owners and simple users.Key Contributions
Within MYRIOPOS, UBITECH drives the design and development of advanced intrusion detection techniques and mechanisms for the identification of illegal electronic content in the web and dark web to ensure operational security in distributed systems of systems.Protected Query & Learning Engine (Software)
The Protected Query & Learning Engine allows for secure computation and processing on encrypted data, eliminating the need for decryption and ensuring the absolute privacy of sensitive information throughout the entire query and learning process. The engine helps to derive learnt insights and queryable result sets without compromising confidentiality.
Technologies: Trino, Zama, Concrete-ML
Research Papers: Iatropoulou, S., Anastasiou, T., Karagiorgou, S., Petrou, P., Alexandrou, D., & Bouras, T. (2023, April). Privacy-preserving Data Federation for Trainable, Queryable and Actionable Data. In2023 IEEE 39th International Conference on Data Engineering Workshops (ICDEW)(pp. 44-48). IEEE.
Projects: MobiSpaces
Automated Simulation and Visualisation Engine (Software)
A simulation engine that streamlines the process of creating and analyzing complex AI pipelines. By automating key stages, from data loading and model configuration to scenario scheduling, execution and result visualization, the Engine empowers non-experts to rapidly gain actionable insights.
Technologies: Mflow
Projects: AURORA
LLM-driven Criminal Network Reconstruction from OSINT data sources (Software)
The software harvests publicly available Open-Source Intelligence (OSINT) data and automatically reconstructs potential criminal networks. By intelligently processing and understanding vast amounts of text-based information, our technology can identify key individuals, relationships, and operational patterns, providing law enforcement and intelligence agencies with valuable insights for investigation and disruption.
Technologies: Transformers (Hugging Face), Neo4J, NetworkX
Projects: AVALANCHE, PRESERVE
Code Quality and Vulnerability Assessment embedded in DevSecOps lifecycle (Software)
Software incorporated in the DevSecOps lifecycle which provides a unified view of both code and application (binaries) health, integrating critical insights from quality analysis and security vulnerability assessments.
Technologies: SonarQube, Trivy
Projects: AppTake
LLM-driven Code Self-healing (Software)
The software prototype combines ESBMC with LLMs to automatically detect, diagnose, and repair code anomalies or errors both at development and deployment phase.
Object Detection under Adversarial Conditions with Explanations (Model)
An advanced object detection AI model engineered to maintain high accuracy and reliability even when subjected to subtle but intentionally crafted adversarial perturbations. It highlights the development of methods to understand the model's reasoning under attack, leading to better insights, refinement, and trust.
Technologies: YOLO8, ART
Research Papers: Anastasiou, T., Pastellas, I., & Karagiorgou, S. (2024, December). Adversarial Explanations for Informed Civilian and Environmental Protection. In 2024 IEEE International Conference on Big Data (BigData) (pp. 2672-2681). IEEE.
Projects: CYGNUS, TALON
Multi-objective Problem Solving through Evolutionary Algorithms
A collection of multi-objective evolutionary algorithms which suggest optimal solutions under competitive conditions.
Technologies: DEAP
Projects: CYGNUS, TALON
Automated Pentesting Methodology
A novel methodology converging OWASP and NIST best practices that leverages intelligent automation in pentesting preparation, execution and post-exploitation to streamline the testing process. By employing a carefully curated suite of specialized tools and custom scripts, we can identify vulnerabilities across Targets of Evaluation based on their context.
Technologies: Langchain / LangGraph, HuggingFace, CAI
Projects: TRITON
AI Model Benchmarking and Explainability Framework
An automated optimization framework which monitors and benchmarks over diverse AI Models and data modalities to learn patterns about computing, cost, energy allocation and performance accuracy. The framework leverages the benchmarking insights to suggest or automatically enforce in real-time (via state persistency) adjustments to AI model configurations, batch sizes, data loading / smart placing strategies, or even hardware allocation to minimize energy consumption while maintaining desired performance levels.
Technologies: MLflow, explainerdashboard
Research Papers: Theodorou, G., Karagiorgou, S., & Kotronis, C. (2024, December). On Energy-aware and Verifiable Benchmarking of Big Data Processing targeting AI Pipelines. In 2024 IEEE International Conference on Big Data (BigData) (pp. 3788-3798). IEEE.
Projects: RAIDO, TALON, PROTEAS
Group Leader
Dr. Sophia Karagiorgou (Head of Group) – Expertise in Big Data and AI Technologies.
Short Bio
Dr. Sophia Karagiorgou holds the position of Senior Research and Innovation (R&I) manager at UBITECH and is a member of the Board of Directors (BoD) at Adra. She leads the Data and AI/ML Systems (DAI) group of UBITECH. Dr. Karagiorgou is distinguished by her proficiency in Process Modelling, Service-Oriented Architecture (SOA), High-performance Data Analytics (HPDA), and numerous programming languages including Python, JAVA, C/C++, and C#. Her expertise extends to the realms of Big Data, Artificial Intelligence, Computer Science, and Databases. Furthermore, she has contributed to the academic community with a publication record exceeding 40 scholarly articles.
[LinkedIn] [Google Scholar] [ORCID]
Key Team Members
Dr. Christos Kotronis (Technical Manager) – Expertise in Model-driven Engineering, SysML, Simulation-based Analysis, AI System Modeling technologies.
Short Bio
Dr. Christos Kotronis is a senior Technical Manager of the Data & AI/ML Systems (DAI) group at UBITECH and a dedicated teaching assistant at Harokopio University of Athens (HUA), where he has served since 2016. He holds a Ph.D. and has completed post-doctoral research in the Department of Informatics and Telematics at Harokopio University. His doctoral research, funded by the General Secretariat for Research and Technology (GSRT) and the Hellenic Foundation for Research and Innovation (HFRI), focused on developing SysML-based, model-driven approaches for generating Quality-of-Service (QoS), performance, cost, and ethical models. With over 20 publications, his work has been applied in real-world case studies across transportation, e-health, AI systems, and sensor networks, emphasizing real-time simulation and the analysis of dynamic system behavior.
[LinkedIn] [Google Scholar] [ORCID]
Dr. Evangelos Kafantaris (Technical Manager) – Expertise in the Design and Deployment of Computational Algorithms, Data Extraction and Machine Learning Pipelines
Short Bio
Dr. Evangelos Kafantaris is a Technical Project Manager of the Data & AI/ML Systems (DAI) group of UBITECH, with a background in Data Science (PhD) and Electronics and Electrical Engineering (MEng) from the University of Edinburgh. Published research includes projects addressing the challenges of data quality and fit-for-purpose multivariate analysis through the design and application of entropy-based algorithms. Further professional experience includes the strategic analysis of IC manufacturing technologies (San Francisco Bay Area, United States), the design of smart-laboratory ETL pipelines (Edinburgh, United Kingdom) and the optimization of cross-domain ML classification architectures (Basel, Switzerland).
[LinkedIn] [Google Scholar] [ORCID]
Theodora Anastasiou (AI Engineer) – Expertise in AI Systems.
Short Bio
Theodora Anastasiou is a Machine Learning Researcher at the Data & AI/ML Systems (DAI) group of UBITECH, specializing in AI research and adversarial machine learning, with a focus on developing robust, secure, and explainable AI/ML systems. She holds a Bachelor's degree in Informatics and Telecommunications from the National and Kapodistrian University of Athens (NKUA) and a Master's degree in Web Engineering from Harokopio University of Athens (HUA). She is currently pursuing an MBA at the University of Limassol (UOL). With a strong foundation in programming, data science, and web development, she brings hands-on experience in designing and deploying AI models, optimizing machine learning pipelines, and ensuring system security and interpretability.
Irem Doken Goymen (Project Manager) – Expertise in AI, Image Processing and Project/Funding Management.
Short Bio
Irem Doken Goymen is a Technical Delivery Manager of the Data & AI/ML Systems (DAI) group at UBITECH with a background of B.Sc. degree in Electrical and Electronics Engineering from Bursa Uludag University. She studied Biomedical Engineering Master's degree at Istanbul Technical University and recently pursuing a second Master's Degree in Big Data and Artificial Intelligence at National Kapodistrian University of Athens. She has specialized in image processing with ML with multiple publications in the field. In 2018, she was recognized by the Chamber of Electrical Engineers for her award-winning project, 'Tracking System Based on Image Processing with Machine Learning' and applied for patent. Thanks to over four years of experience in both the Technology Transfer Office (TTO) and the private sector, she has strong skills in project and funding management, combining technical background with strategic management capabilities. She successfully completed PMP training and is recently a candidate for the Registered Technology Transfer Professional (RTTP), having collected the required credits approved by the Alliance of Technology Transfer Professionals (ATTP).
[LinkedIn] [Google Scholar] [ORCID]
Stavroula Iatropoulou (Software Engineer) – Expertise in Full Stack Web Development, Distributed Systems Integration, System Analysis & Design.
Short Bio
Stavroula Iatropoulou is a Software Engineer at the Data & AI/ML Systems (DAI) group of UBITECH. She holds a Bachelor's degree in Informatics and Telecommunications from the National and Kapodistrian University of Athens (NKUA). Her thesis on the "Fake News Detection on news articles using Machine and Deep Learning Techniques" was implemented with Natural Language Processing methods and Neural Networks. She is currently pursuing a master's degree entitled "Digital Technologies and Smart Infrastructures in Agriculture" at the Agricultural University of Athens and is working on her thesis on "Remote sensing techniques for detecting water and nutrient deficiency in the plant Valerianella locusta grown in a closed hydroponic vertical cultivation system using a multispectral camera". With a solid foundation in Computer Science and, particularly in programming, data science and web development, she has contributed in designing, developing web applications integrated with modern software systems delivering robust software solutions.
[LinkedIn] [Google Scholar] [ORCID]
Ioannis Makris (Software Engineer) – Expertise in Full stack Web Development, System Analysis & Design
Short Bio
Ioannis Makris is a member of the Data & AI/ML Systems (DAI) group at UBITECH. He holds a B.Sc. in Informatics and Telematics from Harokopio University of Athens, where he graduated with honors for his thesis titled "University Management System"—a web application designed to efficiently manage large-scale university operations related to student and course administration. With a solid background in Computer Science, particularly in the analysis, design, and full-stack development of web applications, Ioannis has contributed to numerous projects focused on building robust data processing systems and applications.
Nikolas Maragkos (Software Engineer) – Expertise in Full Stack Web Development.
Short Bio
Nikos Maragkos is a Full Stack Software Engineer with a Bachelor's degree in Informatics from the University of Piraeus. He holds a Master's degree in Management and Technology from the Athens University of Economics and Business (AUEB) and is currently a PhD Candidate in Artificial Intelligence at the University of Piraeus, focusing on user profiling and trend analysis in social media. He has extensive experience working as a software engineer across a wide range of sectors, including digital products, banking, insurance, telecommunications, the public sector, and debt collection. Nikos has a strong foundation in front-end development, paired with comprehensive back-end skills, enabling him to deliver robust, end-to-end software solutions.
George Pantelis (Software Engineer) – Expertise in Full Stack Development, Distributed Systems Integration, Secure Identity Management, and Data-Driven Web Technologies.
Short Bio
George Pantelis is a Full Stack Software Engineer with a strong academic background and a focus on delivering robust, end-to-end software solutions. He holds a Bachelor's degree in Management Science and Technology and a Master's degree in Computer Science, both from the Athens University of Economics and Business (AUEB). His academic foundation combines business insight with advanced technical expertise. Throughout his career, he has been involved in the design, development, and integration of modern software systems, often contributing to technical decision-making and architectural planning.
[LinkedIn] [Google Scholar] [ORCID]
Ioannis Pastellas (AI Engineer) – Expertise in AI Systems.
Short Bio
Ioannis Pastellas is an AI engineer in the Data & AI/ML Systems (DAI) group at UBITECH. With a strong foundation in Computer Science and Artificial Intelligence, Ioannis Pastellas has led projects that combine machine learning, explainable AI, and reinforcement learning to tackle complex challenges across various domains. His work demonstrates a deep commitment to advancing AI research and transforming methods into real-world solutions. He has published research in respected international conferences and journals, contributing to the broader AI community.
[LinkedIn] [Google Scholar] [ORCID]
George Theodorou (AI Engineer) – Expertise in Explainable AI, Energy Aware AI Technologies, MLOPs.
Short Bio
George Theodorou is a Machine Learning Researcher at Ubitech with a strong foundation in both academia and industry. He holds an MSc in Machine Learning from Imperial College London and a BSc in Management Science & Technology from the Athens University of Economics and Business (AUEB). His professional journey includes significant experience in consulting, having worked as a Machine Learning Engineer at the Accenture Athens Center of Excellence, where he played a key role in innovative projects, leveraging his expertise to drive impactful solutions. Currently, he specializes in AI Explainability, AI energy efficiency, and leveraging AI to combat cybercrime. His work has received global recognition, with publications in two major AI conferences: IEEE Big Data and the World Conference on Explainable AI.
[LinkedIn] [Google Scholar] [ORCID]
Recent Highlights
Publications
- Anastasiou, T., Pastellas, I., & Karagiorgou, S. (2024, December). Adversarial Explanations for Informed Civilian and Environmental Protection. In 2024 IEEE International Conference on Big Data (BigData) (pp. 2672-2681). IEEE.
- Theodorou, G., Karagiorgou, S., & Kotronis, C. (2024, December). On Energy-aware and Verifiable Benchmarking of Big Data Processing targeting AI Pipelines. In 2024 IEEE International Conference on Big Data (BigData) (pp. 3788-3798). IEEE.
- Theodorou, G., Karagiorgou, S., Fulignoli, A., & Magri, R. (2024, July). On Explaining and Reasoning About Optical Fiber Link Problems. In World Conference on Explainable Artificial Intelligence (pp. 268-289). Cham: Springer Nature Switzerland.
- Papaioannou, P., Pastellas, I., Tranoris, C., Karagiorgou, S., & Denazis, S. (2024, July). AI-fuelled Dimensioning and Optimal Resource Allocation of 5G/6G Wireless Communication Networks. In 2024 IEEE International Mediterranean Conference on Communications and Networking (MeditCom) (pp. 413-418). IEEE.
Collaboration & Partnerships
Academia & Research:
Industry:For collaboration opportunities and other inquiries, contact us at [email protected]
