ralf: Real-time, Accuracy Aware Feature Store Maintenance

apply(conf) - May '22 - 10 minutes

Feature stores are becoming ubiquitous in real-time model serving systems, however there has been limited work in understanding how features should be maintained over changing data. In this talk, we present ongoing research at the RISELab on streaming feature maintenance that optimizes both resource costs and downstream model accuracy. We introduce a notion of feature store regret to evaluate feature quality of different maintenance policies, and test various policies on real-world time-series data.

Sarah Wooders

PhD Student

UC Berkeley - RISELab

Sarah Wooders is a second year PhD student in UC Berkeley’s RISELab, advised by Joseph Gonzalez and Ion Stoica. Her current work is focused on real-time feature stores. Before Berkeley, she founded Glisten AI, which builds AI to categorize and tag product data and was part of Y Combinator’s W20 batch. Her undergraduate degree is from MIT, where she studied computer science and math and did research at CSAIL in the Supertech Group. While at MIT, she directed Code for Good, helped organize HackMIT, and interned at MemSQL, MobLab, and Bloomberg.