Policy Learning with Weak Signals

2026-10-07

AI Summary

This study addresses the challenge of optimal policy learning in digital experiments characterized by weak signal-to-noise ratios, high-dimensional covariates, and massive datasets. By formulating treatment effect estimation as bounded signal-to-noise ratio observations, we establish that optimal policies are generally unlearnable. To overcome this fundamental limitation, we introduce smoothness assumptions and leverage Gaussian processes alongside minimax theory to derive adaptive policies based on linear smoothers. Our primary contributions include a formal infeasibility theory under weak signals and a corresponding learnable framework. Large-scale experiments conducted at Netflix demonstrate that personalized smoothing policies significantly outperform non-personalized baselines, achieving vanishing welfare regret.

Abstract

Policy learning in digital experimentation faces three challenges: weak signal-to-noise ratios, rich covariate spaces, and massive data volumes. We formalize this regime by modeling treatment-effect estimates from increasingly fine covariate partitions as Gaussian observations with bounded signal-to-noise ratios. We establish that, in general, the optimal treatment policy is not learnable in this setting. Even learning the optimal policy value suffers from impractically slow rates. However, when treatment effects vary smoothly, we derive minimax-adaptive policies based on linear smoothers that achieve vanishing welfare regret. We demonstrate the practical value of our framework by applying it to large-scale real-world experiments at Netflix, showing that personalized linear-smoothing policies can dominate unpersonalized policies even in this challenging empirical setting.

Problem

• Policy Learning
• Weak Signals
• Signal-to-Noise Ratio
• Digital Experimentation
• Personalization

Innovation

• Policy Learning
• Weak Signals
• Minimax-Adaptive Policies
• Linear Smoothers
• Welfare Regret