CS Colloquium
Machine Learning in Production Database Systems
Time and Location:
Oct. 23, 2024 at 2PM; 60 Fifth Avenue, Room 150Speaker:
Vikram NathanLink:
Seminar homepageAbstract:
The database community has seen a flurry of recent work integrating machine learning components into data management systems, but deploying these techniques in production without a human in the loop has seen mixed results. What’s missing when we try to translate research into practice? And how should that inform how we approach ML + Systems research? This talk will attempt to answer these questions through the lens of the speaker’s experience deploying ML-based Intelligent Scaling (named RAIS) at Amazon Redshift, a cloud data warehouse offered by AWS. RAIS automatically and proactively adjusts the capacity of a customer’s Redshift cluster based on workload patterns and predicted query complexity, with the goal of offering a price-performance tradeoff suited to each customer. The talk will cover the design of RAIS, the challenges in designing a production system that relies heavily on imperfect predictions, and new research directions that arise from it.
Notes:
Vikram Nathan is a senior applied scientist at Amazon Web Services (AWS) in the Learned Systems Group (LSG), where he works primarily on Amazon Redshift. LSG is a systems research group that leads efforts to deploy machine learning in several of Amazon’s cloud data management offerings, including RDS, Aurora, and Redshift. Prior to working at Amazon, Vikram received his PhD at MIT, working with Mohammad Alizadeh and Tim Kraska, where his research covered ML + databases and networks. His work has been published at SIGCOMM, NSDI, SIGMOD, and VLDB, among others.
In-person attendance only available to those with active NYU ID cards. Zoom alternatives are also available.