Policy Evaluation under Interference

on YouTube ↗30m

" summary

Compares two concrete interference approaches: a graphon central limit theorem for network interference that gives root-n convergence for the direct effect (variance inflation captured by an explicit Qi term), shows a naive Horvitz–Thompson-style estimator for the indirect effect diverges but becomes valid after removing principal components of the adjacency matrix via PCA, and an equilibrium approach for marketplaces that estimates a marginal response by regressing on small mean-zero payment perturbations to produce a policy gradient. Simulations and theory show orders-of-magnitude improvement and log T vs sqrt T regret.

AI-written summary · about

" chapters (10)

1. 0:00Introduction
2. 1:46Interference
3. 5:31Network Interference
4. 9:03Direct Effect
5. 17:05Agricultural Subsidies
6. 17:55No Network Structure
7. 18:40Concrete Application
8. 20:27Two Approaches
9. 23:48Results
10. 27:45Questions

" related talks

• 49mCausal Inference in Complex Systems: Network Interference, Strategic Agents, and BeyondPanos Toulis · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 33mSimple Yet Efficient Estimators For Network Causal Inference...Christina Yu · Algorithmic Aspects of Causal Inference 2022
• 28mConfounding-Robust Policy Evaluation In Infinite-Horizon Reinforcement LearningAngela Zhou · Algorithmic Aspects of Causal Inference 2022
• 45mInference and Interference in Marketplace ExperimentationHannah Li · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 09mTowards Credible and Effective Data-Driven Decision-MakingAngela Zhou · Meet the Fellows Welcome Event
• 30mCounterfactual Inference For Sequential Experiment DesignRaaz Dwivedi · Learning from Interventions 2022
• 14mInformation, Learning, And Incentive Design For Societal NetworksManxi Wu · Meet the Fellows Welcome Event
• 01h 08mAn Introduction to Potential Outcomes, DAGs and Single-World Intervention GraphsThomas S Richardson · Causality Boot Camp