A web-based solution for federated learning with LLM based automation

Abstract

Federated Learning (FL) presents a promising paradigm for collaborative machine learning across distributed devices. However, its implementation is hindered by the complexity of building reliable communication architectures and the requisite expertise in both machine learning and network programming. This thesis addresses this challenge by introducing a comprehensive solution that facilitates the seamless orchestration of FL tasks while integrating intent-based automation.

In the initial phase, we develop a user-friendly web application supporting the FedAvg algorithm, enabling users to easily configure parameters through an intuitive interface. The backend solution manages communication tasks between the parameter server and edge nodes, incorporating an efficient communication architecture. Additionally, we delve into implementing secondary functionalities like model compression and scheduling algorithms to optimize FL operations.

Building upon this foundation, we explore the frontier of intent-based automation in FL. Leveraging a fine-tuned Language Model (LLM) trained on a tailored dataset, our solution conducts FL tasks in response to high-level prompts provided by users increasing the accessibility of the provided solution.

Through this thesis, we contribute a practical framework for bridging the conceptualization and implementation of FL solutions. By providing a user-friendly interface, efficient communication, and automated task execution, our solution empowers researchers and practitioners to harness the potential of federated learning effectively.