Serverless Computing
Serverless computing offers a paradigm shift in cloud computing, aiming to automate infrastructure management and provide on-demand scalability for applications, particularly those involving machine learning and data analytics. Current research focuses on optimizing performance and cost-effectiveness through techniques like asynchronous training strategies, knowledge distillation for heterogeneous client models, and the application of reinforcement learning for resource allocation and cold-start mitigation. This approach holds significant promise for improving the efficiency and accessibility of various applications, from federated learning and large-scale inference to data-intensive tasks, while addressing challenges related to security, privacy, and resource utilization.
Papers
SPIRT: A Fault-Tolerant and Reliable Peer-to-Peer Serverless ML Training Architecture
Amine Barrak, Mayssa Jaziri, Ranim Trabelsi, Fehmi Jaafar, Fabio Petrillo
Exploring the Impact of Serverless Computing on Peer To Peer Training Machine Learning
Amine Barrak, Ranim Trabelsi, Fehmi Jaafar, Fabio Petrillo