A Minimax Surrogate Loss Approach to Conditional Difference Estimation
This addresses the challenge of causal inference in machine learning for personalized treatment effect estimation, though it appears incremental as it builds on existing frameworks with specific loss improvements.
The paper tackles the problem of estimating personalized treatment effects with binary outcomes by proposing surrogate loss functions that incorporate both treatment and control data, resulting in tighter bounds and a minimax support vector machine formulation that solves a single convex optimization problem.
We present a new machine learning approach to estimate personalized treatment effects in the classical potential outcomes framework with binary outcomes. To overcome the problem that both treatment and control outcomes for the same unit are required for supervised learning, we propose surrogate loss functions that incorporate both treatment and control data. The new surrogates yield tighter bounds than the sum of losses for treatment and control groups. A specific choice of loss function, namely a type of hinge loss, yields a minimax support vector machine formulation. The resulting optimization problem requires the solution to only a single convex optimization problem, incorporating both treatment and control units, and it enables the kernel trick to be used to handle nonlinear (also non-parametric) estimation. Statistical learning bounds are also presented for the framework, and experimental results.