Zwi Altman (Orange Labs)
Title: AI in radio access networks: challenges and open problems
Schedule: Wednesday 9 September 2026, 9:00-10:30 (1.5 hours)
Abstract:
The integration of AI/Machine Learning (AI/ML) into communication networks necessitates the definition of architecture and interfaces to support the operation of AI/ML services, which can be hosted within the network or on external MLOps cloud platforms. The first part of the presentation will focus on the Open Radio Access Network (O-RAN) and its Service Management and Orchestration (SMO) framework. Specifically, O-RAN specifies open interfaces that enable third parties to introduce, among others, intelligent applications empowered by AI/ML capabilities. The second part of the presentation will address recent challenges related to AI/ML in the RAN, such as the integration of Agentic AI, the incorporation of expert knowledge into ML solutions, and issues related to safe learning and transfer learning.
Biography:
Dr. Zwi Altman is a Senior Expert in Future Networks and a Research Referent at Orange Innovation. He received his B.Sc. and M.Sc. degrees in electrical engineering from the Technion-Israel Institute of Technology in 1986 and 1989, respectively, and his Ph.D. in electronics from INPT France in 1994. He was a Post-Doctoral Research Fellow at the University of Illinois at Urbana-Champaign. In 1996, he joined France Telecom R&D, now Orange Innovation, where he has been involved in various projects related to wireless networks, Massive MIMO, quantum communication, self-organizing networks (SON), and Machine Learning. Since 2018, he has served as an O-RAN delegate.
Alexandru Dobrila (Hivenet/Antimatter)
Title: Distributed AI at the Edge
Schedule: Tuesday 8 September 2026, 9:00-10:30 (1.5 hours)
Abstract: The rapid growth of AI is exposing the limits of traditional cloud infrastructure. Beyond compute capacity, AI has become a challenge of energy availability, data sovereignty, privacy, and efficient service delivery. This talk presents an alternative model for AI infrastructure: distributed, edge-native systems that bring compute closer to where energy and users already exist. Drawing on practical experience from Hivenet and Antimatter, we will explore how a network of geographically distributed nodes can be orchestrated into a single logical cloud for AI inference, storage, and compute services. Topics include serving large language models on consumer GPUs, optimizing inference throughput and memory usage, confidential computing and building resilient systems that operate despite node churn and unreliable networks. The talk will also highlight ongoing research challenges in distributed AI, including energy-aware scheduling, sovereign cloud architectures, fault-tolerant distributed training, and scalable coordination across heterogeneous hardware. Attendees will gain insight into how high-performance computing, distributed systems, and sustainable infrastructure are converging to define the next generation of AI platforms.
Biography:
After a PhD in distributed systems focused on scheduling and fault tolerance, Alexandru Dobrila spent 13 years in industry building large-scale infrastructure. At Stellantis, he worked on storage systems and networking for hyperscale datacenters before leading infrastructure teams responsible for network operations across multiple datacenters and globally distributed software teams developing connected vehicle services. In 2024, he joined Hivenet as Director of Research, where he bridges cutting-edge research with the practical challenges of operating distributed compute and storage systems at scale in collaboration with Inria. He is also involved with Antimatter, contributing to the development of energy-aware, sovereign AI infrastructure that combines distributed cloud software, modular datacenters, and flexible energy resources to support the next generation of AI workloads.
Bruno Gaujal (Inria)
Title: Reinforcement learning: from bandits to structured MDPs
Schedule: Thursday 10 September 2026, 11:00-12:30 (1.5 hours), Friday 11 September 2026, 11:00-12:30 (1.5 hours)
Abstract: This lecture will cover fundamentals of reinforcement learning. It will start by introducing stochastic bandits and the UCB algorithm and continue with constructing regret bounds in general Markov decision processes (MDPs). The last part will discuss several cases where the structure of the MDP can be exploited to get better regret bounds and/or simpler learning algorithms.
Biography:
Frédéric Giroire (CNRS)
Title: Distributed inference in the edge-network-cloud continuum
Schedule: Tuesday 8 September 2026, 11:00-12:30 (1.5 hours)
Abstract: The rapid proliferation of artificial intelligence applications and Internet of Things (IoT) devices has created an unprecedented demand for real-time, privacy-preserving, and bandwidth-efficient data processing. Traditional cloud-centric AI inference often struggles to meet stringent latency requirements, while edge-only solutions are frequently bottlenecked by limited computational and energy resources. To address these bottlenecks, distributed inference across the edge-network-cloud continuum has emerged as a critical paradigm, enabling the intelligent partitioning and offloading of computational workloads across available infrastructure tiers. This talk will explore collaborative intelligence frameworks (such as cascade models, early-exist models, model compression and selection, feature compression, model splitting) designed to optimize inference in resource-constrained environments while maintaining high predictive accuracy.
Biography:
Frédéric Giroire currently is a senior research scientist at CNRS, part of the Coati joint team between I3S (CNRS, University Côte d’Azur) laboratory and Inria which he joined in 2008. He is also a chair of the 3IA Côte d’Azur cluster. He heads the COMRED group at I3S. He earned his Ph.D. from the Sorbonne University (Paris 6) in 2006. His industry experience includes 6 months at Sprint Research Labs (California) in 2002 and a year at Intel Research Labs (Berkeley) in 2007, resulting in 3 patents. His research focuses on AI, algorithmic graph theory, combinatorial optimization, and AI for network design, management, sustainability.
Nirupam Gupta (University of Copenhagen)
Title: Robust Machine Learning: A Quest to Learning in Untrusted Environment
Schedule: Thursday 10 September 2026, 9:00-10:30 (1.5 hours), Friday 11 September 2026, 9:00-10:30 (1.5 hours)
Abstract: As machine learning systems scale in both model complexity and data volume, they increasingly rely on distributed algorithms to process massive datasets efficiently. Yet, with this scale comes vulnerability: data from diverse sources may be noisy, corrupted, or adversarial, and distributed computing environments are prone to hardware faults, software bugs, and malicious attacks. If left unaddressed, these issues can significantly undermine the reliability of large-scale learning systems. In this lecture, I will discuss how to design robust learning algorithms that remain reliable in the face of such real-world challenges. Using the example of stochastic gradient descent, a commonly used optimization scheme in ML, I will present recent advances in Byzantine-robust aggregation and examine how robustness interacts with local stochasticity, data heterogeneity and privacy-preservation. I will conclude with some open problems and research directions at the intersection of theory and practice in robust machine learning.
Biography:
Nirupam Gupta is an Assistant Professor in the Department of Computer Science at the University of Copenhagen, Denmark. He previously held postdoctoral positions at École Polytechnique Fédérale de Lausanne (EPFL), Switzerland, and at Georgetown University, USA (2019-2021). His current research is on distributed machine learning (a.k.a. federated learning), with emphasis on security (robustness) and privacy. He has co-authored a book on the principles of robust machine learning. His co-authored paper has received the best paper award at ICDCN. Nirupam received his Ph.D. from the University of Maryland, College Park, USA (2018) and his Bachelor’s degree from Indian Institute of Technology Delhi, India (2013).
Lorenzo Maggi (NVIDIA)
Title: Bayesian optimization: Theory and applications to telecommunications
Schedule: Wednesday 9 September 2026, 11:00-12:30 (1.5 hours)
Abstract: Bayesian optimization (BO) has emerged as a cornerstone for optimizing complex “black-box” systems where i) the function to be optimized and its gradient are unknown ii) convergence speed is critical, since evaluating the function is expensive and/or risky iii) prior knowledge can be leveraged to guide the search. In this seminar, we will cover the basis of BO (parametric, non-parametric, Gaussian processes, acquisition functions), mention recent research trends (high-dimensional BO, conformal prediction, etc.), and present some applications in telecommunication networks, such as energy savings and power control.
Biography:
Lorenzo Maggi is a senior research scientist at NVIDIA, specializing in the convergence of wireless communications and machine learning. Before joining NVIDIA, Lorenzo developed algorithmic solutions for 5G networks at Nokia Bell Labs, focusing on energy efficiency, beamforming, and radiation mitigation. Prior to this, he worked on network routing algorithms at Huawei. Lorenzo holds a Ph.D. in applied mathematics from Eurecom/INRIA, France and a master’s degree in telecommunication engineering from the University of Pavia, Italy. He won the Best Paper Award at WiOpt 2014 and ITC 2017, and received the Nokia Inventor Award in 2022. He maintains a technical blog at www.mlwithouttears.com.
Farnaz Moradi (Ericsson)
Title: The Road to Autonomous Networks: AI/ML Foundations and Frontiers
Schedule: Tuesday 8 September 2026, 14:00-15:30 (1.5 hours)
Abstract: A fully autonomous network deploys, configures, maintains, and retires itself without human intervention. While most operators today remain at low autonomy levels, advances in AI, agentic AI, and network digital twins are accelerating the journey toward zero-touch operation. Achieving this vision requires embedding AI across the entire network lifecycle. The talk discusses research challenges of AI/ML for autonomous networks: handling heterogeneous and uncertain telecom data; building models that accurately predict the impact of actions and adapt as the environment evolves; and designing models that generalize across use cases, reducing the number of models to maintain. We illustrate these challenges through ongoing research and highlight potential directions for the road ahead.
Biography:
Farnaz Moradi is a master researcher in AI at Ericsson Research. She joined Ericsson in 2014 and has since worked with research and development of AI/machine learning solutions across a range of telecom domains, including intelligent network management and automation. In her current role, she focuses on driving research initiatives for autonomous networks. Moradi holds a Ph.D. in computer science and engineering from Chalmers University of Technology in Gothenburg, Sweden.