Optimizing Federated Learning Techniques for Advanced Decentralized AI Systems
Saurabh Sharma, Zohaib Hasan, Vishal Paranjape · International Journal of Innovative Research in Computer and Communication Engineering · 2023
Machine learning (ML) models have become indispensable in extracting insights and promoting innovation across diverse fields due to the emergence of large data. Historically, the creation of ML models has relied on gathering data in a centralized manner, which has raised notable concerns around privacy and security. This is because sensitive data needs to be transferred to a central place. Decentralized machine learning resolves these concerns by allowing the training of machine learning models on numerous devices without centralizing the data. This study explores the intricacies of decentralized machine learning (ML) training, with a focus on methodologies and algorithms that prioritize privacy preservation. Federated Learning (FL) is a fundamental method in decentralized machine learning that enables training models using data from decentralized sources while guaranteeing data secrecy. Local devices in Florida engage in data calculations and solely transmit model updates to a central server. The central server then combines these updates to improve the global model. This paper investigates the difficulties associated with decentralized training, including the burden of communication, the diversity of data and devices, and the potential for adversarial attacks. Additionally, it evaluates existing solutions and suggests novel approaches to enhance the effectiveness and safety of decentralized machine learning frameworks. The proposed method attains a precision of 97.6%, a mean absolute error (MAE) of 0.403, and a root mean square error (RMSE) of 0.203. The results emphasize the capability of distributed machine learning in creating artificial intelligence systems that protect privacy, can be expanded easily, and are dependable.