Kernel Treatment Effects with Adaptively Collected Data
Not on the map. Why?
Abstract (arXiv)
Adaptive experiments improve efficiency by adjusting treatment assignments based on past outcomes, but this adaptivity breaks the i.i.d. assumptions that underpin classical asymptotics. At the same time, many questions of interest are distributional, extending beyond average effects. Kernel treatment effects (KTE) provide a flexible framework by representing interventional outcome distributions in an RKHS and comparing them via kernel distances. We present the first kernel-based framework for distributional inference under adaptive data collection. Our method combines doubly robust RKHS scores with a witness function learned on one fold, and performs inference on a second fold using a projected, sequentially normalized scalar statistic with valid type-I error. Experiments show that the resulting procedure is well calibrated and effective for both mean shifts and higher-moment differences, outperforming adaptive baselines limited to scalar effects.
Key facts. Cited by 1 paper in the map (data through 13 Oct 2026). Builds on 19 of its 68 references that are in the map. First posted on arXiv 11 Oct 2025.
Builds on
19 of 68 references.
References · 19
- 2018Implicit Quantile Networks for Distributional Reinforcement Learningcited by 363
- 2016Optimal Best Arm Identification with Fixed Confidencecited by 235
- 2019Confidence Intervals for Policy Evaluation in Adaptive Experimentscited by 95
- 2020A Measure-Theoretic Approach to Kernel Conditional Mean Embeddingscited by 86
- 2020Inference for Batched Banditscited by 63
- 2022Optimal Rates for Regularized Conditional Mean Embedding Learningcited by 50
- 2021Conditional Distributional Treatment Effect with Kernel Conditional Mean Embeddings and U-Statistic Regressioncited by 40
- 2021Statistical Inference with M-Estimators on Adaptively Collected Datacited by 40
- 2020Kernel Methods for Causal Functions: Dose, Heterogeneous, and Incremental Response Curvescited by 36
- 2018Counterfactual Mean Embeddingscited by 32
- 2021Post-Contextual-Bandit Inferencecited by 32
- 2021Off-Policy Risk Assessment in Contextual Banditscited by 29
7 more
- 2020Dimension-agnostic inference using cross U-statisticscited by 14
- 2022A Permutation-Free Kernel Independence Testcited by 11
- 2023One-Step Estimation of Differentiable Hilbert-Valued Parameterscited by 11
- 2023An Efficient Doubly-Robust Test for the Kernel Treatment Effectcited by 10
- 2023Sequential Counterfactual Risk Minimizationcited by 6
- 2025Doubly-Robust Estimation of Counterfactual Policy Mean Embeddingscited by 3
- 2025Double Debiased Machine Learning for Mediation Analysis with Continuous Treatmentscited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.