How can data-center networks keep pace with distributed AI training? How can operators verify that an AI agent has configured a network correctly? And what will it take for future 6G networks to support reliable sensing and connectivity across terrestrial and satellite links? Cnam’s 17 new student projects tackle these questions alongside practical challenges in cybersecurity, cloud-native telecom infrastructure and energy-efficient learning at the edge. Industry partners from Nokia, Ericsson, Thales, NeuroTel.ai and Smile co-advise projects across the portfolio, alongside tutors from Cnam and Inria.
Multipath Reliable Transport for AI Data Center Networks
Nathnael Tadesse · Tutor — Jérémie Leguay (Nokia)
Compares packet-level multipath transport with conventional forwarding to measure how it affects throughput, packet reordering and resilience to link failures.
Improving LLM Agents for Closed-Loop Network Configuration
Fahad Waseem · Tutor — Jérémie Leguay (Nokia)
Uses NetConfArena to test whether checking the resulting network state helps LLM agents configure routing more reliably.
Replay Attacks on ISAC Systems
Sumayya Bashir · Tutor — Filippo Rebecchi (Ericsson)
Uses simulation to assess how replay attacks affect range and velocity estimates in integrated sensing and communication systems and explores possible mitigations.
World Models for Networks
Norvan Gorgi · Tutor — Cédric Adjih (Inria)
Explores whether a learned world model can predict how wireless multi-hop connectivity changes as devices move or antennas change direction.
Multi-Connectivity Extension for OCUDU
Youcef Bellouche · Tutor — Stefano Taborelli (Cnam)
Extends a simulation platform to study how mobile devices maintain simultaneous terrestrial and satellite connections as users and low-Earth-orbit satellites move.
Agentic AI for Automated Security Operations
Kema Tarun · Tutors — Agathe Blaise and Wafa ben Jaballah (Thales)
Compares agentic AI frameworks and benchmarks their suitability for security operations such as continuous monitoring and alert triage.
Vulnerability Detection Using Generative AI and Large Language Models
Ali Raza · Tutors — Wafa ben Jaballah and Agathe Blaise (Thales)
Evaluates how generative AI can help detect vulnerabilities in source code or flag signs of security weaknesses in network traffic, including in 5G contexts.
Migration from srsRAN to OCUDU for a Multi-UE 5G Testbed
Waleed Wahab · Tutor — Masoud Baharlouei (NeuroTel.ai)
Evaluates the migration of a software-based 5G testbed to OCUDU by comparing compatibility, configuration and automation with multiple emulated devices.
From Kubernetes to Telco Cloud: Evaluation of Sylva and Nephio for Cloud-Native 5G Infrastructure and Orchestration
Fura Yanji Sherpa · Tutor — Masoud Baharlouei (NeuroTel.ai)
Tests what Sylva and Nephio add to Kubernetes for deploying and managing cloud-native 5G network functions.
Energy and Carbon Footprint of Multimodal Federated Learning in Heterogeneous IoT Networks
Piergiorgio Rodilosso · Tutor — Hugo De Oliveira (Cnam)
Measures energy use and estimates carbon emissions across multimodal federated learning architectures to quantify the trade-off between detection performance and resource costs.
Knowledge Distillation for Model Personalization and Missing Modalities in Heterogeneous IoT Networks
Lluís Expósito Montilla · Tutor — Hugo De Oliveira (Cnam)
Tests whether knowledge distillation can personalize anomaly detection models for IoT devices that lack some types of sensor data.
Efficient Anomaly Detection in IoT Using Spiking Neural Networks
Felipe Alvarez Suares · Tutor — Sotiris Skaperas (Cnam)
Benchmarks a spiking neural network against a conventional neural network on IoT sensor data to compare detection accuracy and computational cost.
Implementation and Evaluation of Pipeline Parallelism in SimAI
Mahmoud El Sayed · Tutor — Davide Avesani (Cnam)
Adds pipeline parallelism to SimAI to study how scheduling, micro-batching and network conditions affect distributed LLM training.
Impact of AI Workloads and Network Topologies on Distributed Training Performance with SimAI
Abdul Hadi Farooq · Tutor — Davide Avesani (Cnam)
Uses SimAI to compare how different AI workloads and network topologies affect training time, link utilization and congestion.
Efficient Retraining of Machine Learning Models at the Edge
Melat Worku · Tutor — Sotiris Skaperas (Cnam)
Tests whether retraining when prediction errors drift can maintain accurate resource-demand forecasts while reducing the cost of model updates.
Predicting 5G Handover Dynamics
Nathnael Bekele · Tutor — Vania Conan (Cnam)
Uses a dataset from a research project to forecast 5G handovers one hour ahead across the Greater Paris region.
Evaluating Physical Property Transfer in 3D JavaScript Simulations
Namus Hulkero · Tutor — Jonathan Rivalan (Smile)
Develops metrics to assess how faithfully physical properties inferred from real-world footage transfer to a JavaScript physics engine.
As AI clusters scale and 5G platforms evolve toward 6G, these projects give students a chance to test design choices while the technologies are still taking shape. Co-advising by industry partners across the portfolio brings operational questions into that work, from what happens when a network path fails to how much energy an edge model consumes. Students will investigate those questions through prototypes, testbeds, simulations and real-world data.

