Rise of the Machines: Removing the Human-in-the-Loop
summary
System automatically tunes linear-combination ranking hyperparameters (example: in-app threshold x_th combining p_click and p_visit) as a constrained black-box optimization: maximize sessions while keeping clicks >= control and send volume <= control. It models each utility with Gaussian processes, runs Thompson sampling to sample posterior maximizers, and updates an hourly parameter-distribution store used for online hashing-based serving. Deployed across notifications, feed, ads and People You May Know, it replaces months of manual A/B tuning and accelerates experimentation.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:30Agenda
- 1:56LinkedIn
- 2:16Notifications
- 3:50Business Metrics
- 4:33Ranking Function
- 5:52Challenges
- 6:30Reformulation
- 8:27Blackbox Optimization
- 10:50Thompson Sampling
- 11:43System Architecture
- 16:11Results
- 19:11Future Directions
related talks
21mDemystifying Machine Learning for Programmatic UA & Retargeting
15mAI-Driven Application Optimization
01h 01mContinuous Optimization of MIcroservices using Machine Learning
41mThompson sampling for personalising a car brands advertisements
18mSynthesizing human and machine capabilities
20mWinning the Game: Leveraging Contextual Bandits for Creative Optimization
37mSupercharging CTR (Click-Through Rate) with Plug-and-Play AI Capabilities
13mAlgorithmic Transparency in Personalized Advertising: Learnings from Building "Why Am I