There is growing recognition that AI/ML techniques will be transformative for innovations in future connected, collaborative compute (3C) networks, not only for the creation of Digital Twins but also for many of the control functions and algorithms underlying, e.g., transport protocols, routing, switching, traffic engineering, radio access, and more. AI-based models provide the basis for new and ever-faster innovations across existing and emerging protocols, management platforms, and physical infrastructures, including datacom environments such as IP/MPLS backbones, Ethernet fabrics, data-center interconnects, cloud networks, radio, optical, and others. The basis for any such AI-driven models, and thus for advances in mechanisms, is the growing need for operational and system-relevant data to train them. Operational network telemetry capabilities provided by new-generation network protocols offer powerful tools and programmable monitoring functions for transport, edge, cloud, and Datacom resources, enabling the collection of inherently heterogeneous real-world data. With 6G penetration across many verticals of the Digital Society, unlocking suitable data within those vertical communication environments is key to driving innovations tailored to the often-varying needs and constraints of those verticals.
Most of the ICT community’s effort is devoted to developing innovative techniques to efficiently transfer knowledge across structurally and temporally distinct domains. Current network-AI approaches often assume a static topology, a homogeneous monitoring stack, and a single administrative domain. In practice, however, operational datacom domains evolve continuously, feature spaces are partially incompatible across packet, optical, edge, and cloud layers, and the cost of rebuilding models from scratch is prohibitive. The ICT community is addressing these problems by treating each operational stakeholder domain as a contributor to a shared learning process.
A key point of the proposed research activities on transfer learning innovation is that it turns heterogeneous datasets into reusable AI assets. It will reduce the amount of labeled data required for a new deployment, shorten task calibration time, and enable reusable model services offered through AIaaS interfaces. The ambition is therefore both methodological and industrial: to establish a credible route from fragmented network and datacom datasets to portable AI models that remain effective under realistic domain shift.
Addressed Key Questions:
Series organizers:
Dirk Trossen, DaPaDOT Tech
Pietro Cassarà, ISTI-CNR
Prof. Giuseppe Aceto, University of Naples Federico II
Luigi Iannone, Huawei
Upcoming Sessions
Date: 24 September 2026, 15:00 CEST
Title: Towards AIaaS for the Networks of the Future: Reusable and Interoperable Data-based Models in Railway Scenarios
Invited Speaker: Alessandro Capuozzo
Organisation: Hitachi Railways, Italy
Web Link: Click Here
Bio
Alessandro Capuozzo is Innovation Manager and Head of AI at Hitachi Rail GTS in Italy. With more than 18 years of experience in embedded software, system architecture and artificial intelligence, he leads the development of AI-based solutions for transportation and industrial applications. His work focuses on bringing AI into real-world systems by combining sensors, perception, reasoning and decision-making capabilities. His areas of expertise include Edge AI, Computer Vision, Vision-Language Models, autonomous systems and railway technologies. He is particularly interested in the evolution of Physical AI and the role of AI in future mobility and industrial automation.
Abstract
The talk explores the evolution from isolated AI models and datasets toward AI-as-a-Service (AIaaS) ecosystems for networking and industrial systems. Particular attention is given to transfer learning, model portability, and the emergence of Physical AI, where intelligent models move beyond data processing to interact with real-world assets, sensors, and autonomous systems.
From railway infrastructures and industrial automation to autonomous vehicles and edge computing platforms, AI models are becoming reusable industrial assets that can be deployed, adapted, and orchestrated across heterogeneous environments. The session will discuss how AIaaS can accelerate this transformation by enabling interoperable AI services while preserving scalability, explainability, and operational reliability.
The presentation outlines a methodological and industrial roadmap toward an ecosystem of distributed AI capabilities, where machine perception, reasoning, and decision-making can be continuously transferred across domains and integrated into physical systems operating in real time.
Date: 9 October 2026, 10:00 CEST
Title: From Heterogeneous Datacom Paradigms to Portable AI Models for 6G Networks
Invited Speaker: Riccardo Burrai
Organisation: SIRIUS Technology, Italy
Web Link: Click Here
Bio
Riccardo Burrai is a network architect and entrepreneur, CTO and Co-Founder of Sirius Technology. He specializes in advanced architectures, with a focus on SRv6 and eVPN. His background is in designing scalable and distributed infrastructure for complex environments, including multi-domain networks. He focuses on topics such as networking, automation and IoT systems, extending control to the edge applied to solutions such as intelligent video surveillance, with an approach oriented towards innovation, efficiency and scalability. In 2007, he co-founded Sirius Technology. Initially focused on creating the radio network in Tuscany, since 2016 the company has extended coverage to Sardinia, using cutting-edge technologies such as artificial intelligence, IoT sensors, and next-generation protocols over IPv6. Sirius’ intelligent infrastructure involves link degradation and saturation of radio and fiber optic bindings, anticipating traffic engineering with the use of the SRv6 protocol. Sirius is not limited to WiFi, but has a Business Unit dedicated to the design of fiber access networks, wiring entire industrial areas and towns in both Sardinia and Tuscany.
Abstract
Future 6G networks will integrate heterogeneous domains spanning radio access, IP and optical transport, edge, cloud, and satellite infrastructures. While each domain already generates large volumes of operational data, differences in telemetry models, protocols, semantics, and timescales make it difficult to exploit this information as a common foundation for Artificial Intelligence.
This talk explores how programmable telemetry, IPv6 and Segment Routing architectures, and next-generation network protocols can transform operational data into structured and reusable assets for AI training and inference. Drawing from real network deployments, we will discuss how telemetry collected across routing, transport, cloud, radio, and optical systems can support portable AI models capable of operating across different network domains.
The objective is to move beyond traditional monitoring toward AI-native networking, where models can understand network state, correlate cross-domain events, predict anomalies, optimize resources, and ultimately contribute to autonomous network operations and new 6G services.
Date: 23 October 2026, 15:00-16:00 CEST
Title: Transfer Learning for Dynamic and Multi-Domain for Future Datacom Services
Invited Speaker: Pietro Cassarà
Organisation: ISTI-CNR, Italy
Web Link: Click Here
Bio
Dr. Pietro Cassarà is a Senior Researcher at the Institute of Information Science and Technology (ISTI) of the National Research Council (CNR). His research focuses on analysing and designing AI-based control and management systems for telecommunications networks. In this field, he is in charge of the network infrastructure
of the joint ISTI-Huawei laboratory at the ISTI. Dr. Cassarà has collaborated with international institutions such as NATO’s Centre for Maritime Research and Experimentation (CMRE) as a temporary research staff member, working on intelligent routing protocols for underwater wireless networks. He has also coordinated project activities in national and European projects on AI-driven network management and has led Proof of Concept (PoC) initiatives
within ETSI, exploring the application of the SRv6 protocol alongside AI-based strategies for IPRAN management. He is a co-author of the IETF Internet draft Generative AI for Intent-Based Networking. He is a founding member of the DIA with the ISTI-CNR. Beyond his research activities, Dr. Cassarà serves as an editor for journals and actively contributes to the organization of conferences, as well as teaching activities in Computer Science and Applied
Mathematics at the University of Pisa.
Abstract
Future datacom services will operate across heterogeneous technology domains, including infrastructures whose topology and ownership evolve continuously. This presentation examines transfer learning as the mechanism that converts such heterogeneous information into reusable AI assets. It explains why conventional network AI, built around static topologies, homogeneous monitoring stacks, and single administrative domains, loses effectiveness in the face of feature mismatch, temporal drift, configuration changes, and cross-layer dependencies. The talk addresses the need for shared representations, domain-specific adapters, topology-aware learning, and self-supervised pretraining approaches. Together, these techniques reduce labeled-data requirements, shorten adaptation time, and limit the cost of rebuilding models for each deployment. In particular, it becomes fundamental to define frameworks that support deciding which knowledge should remain invariant, which parameters should be adapted locally, and which information must stay private. Finally, the talk proposes AI-as-a-Service pipelines, portability metrics, realistic cross-domain benchmarks, and incentive mechanisms that enable operational stakeholders to collaborate without surrendering control of sensitive data. The resulting roadmap links methodological research with deployable, trustworthy, and economically credible 6G-era network operations at scale across diverse industrial verticals.
Keywords: transfer learning; domain shift; datacom; network AI; 6G; AIaaS.
Date: TBA (mid November)
Title: Explain, Optimize, Defend: The Multiple Roles of XAI in Data Communications
Invited Speaker: Prof. Giuseppe Aceto
Organisation: University of Naples Federico II, Italy
Bio
Giuseppe Aceto is Associate Professor at the University of Napoli Federico II. He has a PhD in telecommunication engineering from the same University, where he teaches Bachelor and Master courses in computer engineering, about computer programming and decentralized applications. His work falls in monitoring network performance and security. His recent research focuses on advanced Deep Learning architectures for traffic classification and prediction, on which he has co-authored highly-cited papers in top-notch international journals. He is the recipient of several awards including the 2018 Best Journal Paper Award by IEEE CSIM, and the Computer Networks 2020 Best Paper Award.
Abstract
Artificial intelligence is increasingly used for traffic classification, intrusion detection, and automated network management, but the opacity of deep learning models remains a barrier to trustworthy industrial deployment. This talk explores how eXplainable Artificial Intelligence (XAI) can support multiple aspects of the lifecycle of AI-enabled datacom solutions. Drawing on applications in mobile traffic classification, incremental learning, and intrusion detection, it shows how explanations can reveal model behaviour, validate the relevance of traffic features, and compare alternative architectures. XAI can also guide model simplification and input reduction, enabling faster processing, lower computational costs, shorter traffic observation windows, and earlier attack detection. However, explainability also creates security risks: information about influential features can help adversaries construct effective evasion attacks. The talk therefore provides an overview touching the multiple facets of XAI as an enabler for more transparent, efficient, adaptive, and secure network intelligence.
Date: TBA (early December – first week)
Title: Huawei insights on AI and the datacom sector
Invited Speaker: Dario Rossi, PhD
Organisation: Huawei Paris Research Center, France
Bio
Dario Rossi is Vice President of the Huawei Research Center in Paris, France, and Director of Huawei AI4NET Lab for Network Products & Solutions. He holds an HDR from UPMC, and PhD and MSc degrees from Politecnico di Torino. Before joining Huawei in 2018, he held Full Professor positions at Telecom Paris and Ecole Polytechnique and was holder of Cisco’s Chair NewNet@Paris. He chaired and participated in the steering and technical program committees of 50+ conferences in the networking (incl. SIGCOMM, CoNEXT, INFOCOM) and AI (incl. ICLR, NeurIPS, AAAI) domains and in the editorial board of 5+ journals. He has coauthored over 25 patents and 250 papers in leading conferences and journals, that attracted about 10k citations for an H-index of 50+ and received 10 best paper awards. He is the recipient of a Google Faculty Research Award (2015), an IRTF Applied Network Research Prize (2016) and several Distinguished Member recognition of the IEEE INFOCOM TPC. He is a Member of CLAIRE, Senior Member of IEEE and a Distinguished Member of ACM for contributions to AI and Networking.
Abstract
To increase generalization capabilities of an agent in an automated fashion: it is necessary to minimize human intervention in the agentic evolution loop: rather, human effort is needed to define benchmarks and evaluations practices, and venues like the DIA are the right forum for vendors, ISPs, OTT, academia & power users to discuss and agree on useful realistic benchmarks. However, instead of discussing only the shiny aspects of agents, it is important to analyze as well the dark side of their evaluation, i.e., when things can go wrong and benchmarking creates illusion of progress, or silent error and oddities plague the agent execution. For all these cases, community and benchmarks should explicitly track, count and expose failure modes to increase trust and ultimately foster adoption.
