arXiv · 2611.12287
Algorithm-aware Statistical Inference in High-dimensional Estimating Equations
Abstract
Iterative algorithms are integral to many statistical procedures, whereas sampling uncertainty along their iteration is generally not quantified. We develop simultaneous inference for discrete trajectories generated from high-dimensional estimating equations, with iteration-specific population estimands that preserve algorithmic regularization. Our proposed fixed-Jacobian bootstrap propagates perturbations along the observed trajectory without repeatedly running the original nonlinear algorithm. We establish uniform first-order expansions and bootstrap validity over possibly infinite sets of contrasts and iterations under stability, entropy and moment conditions. Simulations in sparse linear regression, paired Fisher discrimination, and temporal-difference learning demonstrate accurate simultaneous coverage. A micro-randomized trial application illustrates inference for a treatment trajectory with participant-level clustering.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Tongyu Li, Alexander Giessing. 2026-11-09. Algorithm-aware Statistical Inference in High-dimensional Estimating Equations. https://arxiv.science/abs/2611.12287
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.