Orthogonal Statistical Learning with Self-Concordant Loss
This work provides incremental improvements for researchers in statistical learning and causal inference, focusing on two-stage prediction methods.
The paper tackles the problem of improving non-asymptotic excess risk bounds in orthogonal statistical learning by using self-concordant loss functions, resulting in bounds that remove the strong convexity assumption and reduce dimension factors.
Orthogonal statistical learning and double machine learning have emerged as general frameworks for two-stage statistical prediction in the presence of a nuisance component. We establish non-asymptotic bounds on the excess risk of orthogonal statistical learning methods with a loss function satisfying a self-concordance property. Our bounds improve upon existing bounds by a dimension factor while lifting the assumption of strong convexity. We illustrate the results with examples from multiple treatment effect estimation and generalized partially linear modeling.