Gaussian process bandits proposed for constrained black-box lever optimization
Original title: Gaussian Process Bandits for Constraint Black Box Lever Optimization
Authors report a statistically significant efficiency gain versus epsilon-greedy in one live production randomized trial.
The paper proposes Gaussian Process Upper Confidence Bound methods to allocate shared budget weights when feedback is delayed, noisy, and aggregate and adjustments are infrequent. A synthetic simulator tests kernel choices under stationary, drifting, and reactive conditions and identifies a composite Matérn plus Linear kernel. The authors say a large production randomized controlled trial showed a statistically significant efficiency lift over an epsilon-greedy baseline.
Unconfirmed: The efficiency lift is the authors' result from one production trial; the magnitude is not stated.
- Classical independent multi-armed bandits are called inefficient under sparse feedback.
- The Matérn plus Linear kernel is reported as the most robust tested choice.
- Validation is a live production randomized controlled trial against epsilon-greedy.
- The method is described as segmentation-agnostic and shares information across neighboring configurations.
From the original article
1 min read at arxiv.orgOptimizing discrete allocation weights across competing segments of a shared budget is a fundamental challenge in resource-constrained, sequential decision-making environments. In practice, this optimization is often hindered by delayed, noisy, and aggregate feedback, alongside strict operational constraints on how frequently weights can be adjusted.
By Bram van den Akker · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.