If you wish to participate simply fill up the contribution form.
Over 345 million Spotify users rely on Spotify’s great recommendations and personalized features in 170 different markets around the globe (with 85 of these markets launching in the first part of 2021. With the massive inflows of data and complexity of production use cases, defining a unified approach to ML is challenging. In this talk, Aman and Martin give an overview of the challenges we face with building a central ML Platform at a highly autonomous organization, and our approach of adoption by incentive. They dive deeper into our history with feature tooling, the foundation they are building now and where they are headed with a Feature Marketplace strategy at Spotify.
Fabio Buso from Logical Clocks discusses how to leverage the online feature store can to compute and ingest features and make them available to operational models making real-time predictions, with low latency and preventing skew between the training and serving features. Arbaz Khan talked how Doordash nearly tripled our cost reduction by employing Redis Hashes efficiently and caring for how we serialize different kinds of feature data.
Davor Bonaci and Max Boyd from Kaskada discuss the importance and challenges of using event-based and time-series data in feature engineering. Simarpal Khaira and Ayan-Anwar Habeeb will introduce Intuit's novel work in feature engineering.
Simba from Featureform describers how a virtual feature store can provide all the value of a traditional feature store without having to re-write a single feature or dealing with vendor lock-in.
Disclaimer: This meetup included a presentation from Airbnb's Zipline feature store. Due to internal policies the company has requested to be removed from the recording.
Atindriyo Sanyal, Technical Lead at Uber, discusses how feature engineering is done at Uber with the first feature store, Michelangelo. Monte Zweben, CEO at Splice Machine, presents the the motivation and key differences in the company's ML architecture and new approach for a Feature Store, with a single engine.