- Balancing weights for dynamic treatments
- Conformal prediction methods for surveys (and, by extension, design-based causal inference)
- Bandit experimentation / best arm identification with non-stationarity (i.e. arm rewards change over time)
- Empirical bayes with heteroskedasticity
- double-robustness under Ill-posedness (for estimators characterised as solutions to an inverse problem, e.g. proxies or noparametrics IV)
- Adaptive Debiased Machine Learning (this literature continues to astound with the platitudes that can be stuck in front of machine learning) - start with sec 2.1
- Learning a basis for nonparametric regression with Hermite Polynomials
- R-learner for heterogeneous effects as weighted pseudo-outcome regression