ABSTRACT
Super Learning is a generalized loss-based ensemble learning framework that was theoretically validated in [42]. This template for learning is applicable to and currently being used across a wide class of problems including problems involving biased sampling, missingness, and censoring. It can be used to estimate marginal densities, conditional densities, conditional hazards, conditional means, conditional medians, conditional quantiles, conditional survival functions, among others [43]. Some applications of Super Learning include the estimation of propensity scores, dose-response functions [21], and optimal dynamic treatment rules [29], for example.
