Application of Online Resource Allocation Based on Ray
summary
Ant Group deploys an online resource allocation system built on Ray for coupon, recommendation and ad allocation that targets maximizing platform revenue under limited inventory. The pipeline uses two-stage: ML scoring then operational-research LP/integer programming, real-time planning algorithms that estimate future traffic and compute dual variables, and an online rank service for millisecond decisions. Implementation leverages Ray placement groups and heterogeneous acceleration; experiments report single TPU solves optimization with hundreds of constraints in seconds and GPU runs that speed algorithms by orders of magnitude.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:46Online resource allocation definition
- 5:42Challenges in solving LP problems
- 7:10Two-stage remodeling scheme
- 11:04Implementation details
- 14:56Ray advantages for system
- 16:44Conclusion and future work
related talks
26mUsing Ray On Large-scale Applications at Ant Group
42mRay: A distributed system for emerging AI applications
22mRay, a Unified Distributed Framework for the Modern AI Stack
32mHow Uber Optimize Marketplaces with Ray
29mAdopting Ray @ LinkedIn
30mTransforming Grab’s AI Platform with Ray
29mScaling ML Pipelines for Autonomous Driving with Ray
10mKeynote: Building a Fusion Engine with Ray