Researcher Pisa

Researcher
Pisa
Toscana, Pisa

PHD STUDENT/POST-DOCThe Ubiquitous Internet Unit of IIT-CNR (Pisa, Italy) is scouting for talented researchers (PhD and post-doc level).

We are collecting expressions of interest for the following research topic:Decentralized Federated Learning in Dynamic SystemsPositions are available at both the PhD and postdoctoral levels, with research directions tailored to the applicant's expertise.

The work may involve a more algorithmic approach, focusing on designing adaptive and efficient learning techniques, or a more theoretical/mathematical perspective, modeling the underlying dynamics of decentralized learning. Candidate profileMSc or PhD in Computer Science, Mathematics, or Physics (complex systems)Background or strong interest in machine learning, decentralized systems, or network scienceProficiency in programming (e. g. , Python, ML frameworks)IIT-CNR will open a formal call for the position.

The scouting process is intended to advertise the above topic in view of the call.

On the research topicA new wave of AI is emerging, characterized by increasing decentralization and the need for federated solutions without central controllers.

As data scales exponentially and becomes more pervasive, concerns about confidentiality, robustness, and resource efficiency demand a shift away from centralized architectures.

Decentralized Federated Learning (DFL) enables user devices, IoT elements, and embedded systems to collaborate without relying on a central authority, enhancing flexibility, resilience, and security. Moreover, as AI progresses, data is becoming a critical bottleneck—most public sources have already been extensively mined, making it essential to extract knowledge from local, private datasets while preserving privacy.

The ability to leverage decentralized learning while ensuring data remains on-device will be a cornerstone of next-generation AI solutions. So far, most research on DFL has focused on static environments, where devices interact with the same set of nodes over time and data never changes.

However, real-world decentralized systems are constantly changing—connections appear and disappear, data shifts, and devices come and go.

This dynamic nature introduces major challenges that have been largely unexplored, such as:Are standard DFL models robust to changing conditions? How can we optimize learning in constantly shifting networks? How can devices effectively share knowledge when neighbors keep changing? How can we learn from new data without forgetting past information? To tackle these questions, DFL in dynamic environments needs to incorporate ideas from continual learning, ensuring that AI models can adapt and improve over time without losing valuable knowledge. Funding and partnershipsThe activities of this topic will be supported by FAIR: Extended Partnership on Artificial Intelligence (funded by the National Recovery and Resilience Plan (NRRP), European Union - NextGenerationEU). Further information: *