causal inference
- Optimal Conditional Inference in Adaptive Experiments: Inference using the last batch only is impossible to improve upon for a wide class of designs
- Adaptive Neyman Allocation: Efficiency gains from neyman allocation in multi-stage experiments
- see also 2-stage exposition from Hahn et al
- Covariate Adjustment in Stratified Experiments: improving upon the interacted regression
- Forecasted Treatment Effects for panel settings with no untreated units
- Choosing a proxy metric from past experiments as a portfolio optimization problem
- Regression estimators for causal effects in network experiments with approximate neighbourhood interference
ML
- Kernel Regularized Least Squares (KRLS) as a hierarchical model and made tractable using random sketching
- Foster-Warmuth Regression: better basis regressions for outcome modelling (and hence CATE estimation)
- Debiased LASSO with general gaussian designs
- Debiasing regularized regression estimators without well behaved covariance matrix
- Tractable sum of squares approximations for \(f-\) divergences
- Optimal Subgroup selection: identifying a region of the feature space where the regression function exceeds a threshold (wlog 0) as a constrained optimisation problem
- Review / intellectual history of computational tooling for bayesian inference