Efficient and Secure Deep Learning Asynchronous Training in Serverless P2P Networks

In this work, we leverage the power of AWS Lambda and S3 buckets to propose a revolutionary method for peer-to- peer deep learning. Unlike traditional synchronous approaches, ours adopts asynchronous training, enabling every peer to advance independently. By guaranteeing the integrity and confidentiality of transferred data, encryption techniques improve data security during communication. By reducing data transfer overhead and maximizing parallelism, dataset partitioning makes effective model training possible. The peer-to-peer deep learning landscape is redefined by this decentralized, secure, and scalable architecture, which promises significant gains in efficiency and model convergence.