ECRP: Large-scale Deep Learning for Intelligent Networks

Project Details

• Funded by ASCR, Networking for Science
• Program manager: Thomas Ndousse
• Early Career Research Program (start: 2017)
• PI: Mariam Kiran (LBNL)
• Postdoc: Being interviewed

Problem

Achieving better resource utilization, predicting failures and improving usage for high-demand science workflows

Applications

Real-time science workflows and heavy traffic demands in astrophysics, bioinformatics

Algorithms

• Lightweight real-time deep learning
• Multi-agent reinforcement learning for optimizing distributed resources

Significance and Impact

Building real-time deep learning algorithms for networks to self-heal and reconfigure, imperative to advance successful science workflows