Ongoing

• Antonio Napolitano · Emergent Failures in Multi-Agent LLM Systems

  09/2026 - ongoing

  Supervisors: Patrizio Dazzi, Jacopo Massa, Alessandro Berti

  University of Pisa | Computer Science

  Abstract : Antonio Napolitano
  The thesis develops and evaluates "emersight" , a tool for observing AI agents that communicate, use tools and share memory. It records actions and memory accesses, reconstructs their relationships, and tests how these traces support interaction reconstruction and diagnosis in controlled delegation and shared-state experiments across LangGraph and AG2.

• Dieudonne Iyabivuze · Personalized Model Poisoning Attacks in Decentralized Federated Learning

  09/2026 - ongoing

  Supervisors: Stefano Chessa, Patrizio Dazzi, Jacopo Massa

  University of Pisa | Computer Science

  Abstract : Dieudonne Iyabivuze
  The thesis evaluates BALANCE, a similarity-based defence for decentralised federated learning, against personalised and non-personalised poisoning. It compares how the defence handles malicious model updates when an attacker sends the same poisoned parameters to every neighbour or tailors a different update to each one.

• Gabriele Denti · Data-aware scheduling in Kubernetes

  12/2025 - ongoing

  Supervisors: Jacopo Massa, Stefano Forti, Patrizio Dazzi, Antonio Brogi

  University of Pisa | Computer Science

  Abstract : Gabriele Denti
  The thesis integrates DA-Placer as a scheduler/controller for Kubernetes to make workload placement aware of data characteristics exchanged across the network. It evaluates the resulting controller against Kubernetes' standard scheduler, assessing whether data-aware decisions improve the performance and efficiency of distributed applications.

• Davide Orsucci · ECLYPSE-NET: Extending Cloud-Edge Simulation with Network Modeling

  12/2025 - ongoing

  Supervisors: Jacopo Massa, Stefano Forti, Patrizio Dazzi, Antonio Brogi

  University of Pisa | Computer Science

  Abstract : Davide Orsucci
  The thesis extends the ECLYPSE simulator to include an advanced network model, enabling the modelling of more realistic and complex Cloud-Edge scenarios and improving the analysis of the performance and efficiency of distributed applications.