Marius: Machine Learning over Billion-Edge Graphs - 10x Faster and 5x Cheaper

Explore a groundbreaking software system designed to revolutionize machine learning on massive graph datasets. Delve into Marius, presented by Assistant Professor Theo Rekatsinas from the University of Wisconsin-Madison, which tackles the critical challenge of data movement during training. Discover how Marius employs an innovative data flow architecture to maximize resource utilization across the entire memory hierarchy, including disk, CPU, and GPU memory. Learn about the system's no-code paradigm, allowing users to define models and benefit from resource-optimized training without complex coding. Understand how Marius achieves remarkable performance improvements, training deep learning models on graphs with over a billion edges and 550GB of total parameters 10 times faster and 5 times more cost-effectively than leading industrial systems. This 31-minute talk from the MLOps World: Machine Learning in Production conference offers valuable insights for researchers and practitioners working with large-scale graph data and machine learning applications.