Swarm Learning ~ Distributed ML for Edge Adaptive Computing
DescriptionMachine learning (ML) has enjoyed success in applications like sensor detection and object identification . In a centralized ML approach, the training data is aggregated in a centralized location, where machine learning models are developed, trained, and tested. However, the centralized approach is facing mounting technical and socioeconomic challenges. Data sovereignty, security, and privacy can all create barriers to transferring and aggregating the vast amount of data required to train ML models. In addition, the costs and carbon impact of moving massive amounts of data can be prohibitive. The industry is calling for an alternative. The alternative should adapt to, and take advantage of, the increasingly distributed nature of data. It should achieve comparable accuracy to centralized learning but outperform centralized learning in terms of security, fault-tolerance, and latency. Swarm Learning is a decentralized ML solution utilizing computing power near the distributed data sources with the proven security of blockchain. In Swarm Learning, both training of the model and inferencing with the trained model occur at edge, where data is fresh, and prompt data-driven decisions can be made. Only learned insights instead of the raw data are shared among collaborating ML peers, which tremendously enhances data security and privacy.
TimeWednesday, June 2811:30 - 12:00 CEST
Computer Science, Machine Learning, and Applied Mathematics