The University of Nicosia (UNIC) Artificial Intelligence Lab (AILab) is pleased to announce the award of a new funded project supported by the NEOLAIA University Alliance (https://neolaiacampus.eu/) seed grant scheme.

The project is titled “LLMs4FL: AI-Assisted Prototyping of Federated Learning Systems: An Empirical Evaluation of Multi-LLM Code Generation, Software Quality, and Token Economics”. It aims to advance the intersection of distributed systems, machine learning, and empirical software engineering.

A Cross-Institutional Collaboration

Principal Investigator Dr. Demetris Trihinas (UNIC AILab) leads the LLMs4FL project in close collaboration with Dr. Panagiota Chatzipetrou from Örebro University (Center for Empirical Research on Information Systems). The core research team is further strengthened by the expertise of Prof. Ioanna Dionysiou from the UNIC Information Security Lab, alongside dedicated doctoral researchers from both partner institutions.

Advancing Federated Learning Prototyping

Aligning directly with the NEOLAIA Digital Transformation pillar, the project addresses the significant complexities associated with developing Federated Learning (FL) systems. FL is an essential AI paradigm that enables collaborative model training across distributed devices while keeping raw data localized. This privacy-preserving approach is highly relevant for sensitive domains (including healthcare diagnostics, financial fraud detection, and autonomous AI-IoT systems).

Despite its advantages, prototyping FL systems requires developers to manage distributed computing challenges and tune numerous configuration parameters. To overcome this severe coding bottleneck, the LLMs4FL project will leverage Large Language Models (LLMs) via the EVOLVE AI platform to generate FL codebases automatically.

The research will evaluate the baseline capabilities of multiple LLMs and examine the impact of personalization techniques. By combining retrieval-augmented generation (RAG) with structured in-context prompting over a curated repository of FL prototypes, the team will assess whether personalized AI models can better understand distributed structures, identify architectural errors, and improve final code quality.

Key Project Objectives

The LLMs4FL project will push beyond existing benchmarks that focus on isolated programs by evaluating generative AI in the context of multi-node distributed systems. The research contributions will focus on three main areas:

Code Generation: Evaluating how effectively generative AI handles complex, multi-component machine learning architectures.

Software Quality: Assessing the functional correctness and long-term maintainability of AI-generated codebases to ensure that human developers and AI agents can safely modify or extend the architecture.

Token Economics: Developing a standardized methodology to compute the exact token expenditure and monetary cost required to generate a fully functioning FL pipeline.

For more updates on the LLMs4FL project and other ongoing research, please continue to follow the UNIC AILab news section.