Kernel Debiased Plug-in Estimation: Simultaneous Debiasing Without Influence Functions
Ivana Malenica , PhD
Assistant Professor -Department of Biostatistics| University of North Carolina
WHEN: Wednesday, November 18, 2026, from 3:30 to 4:30 p.m.
WHERE: Hybrid | 2001 McGill College Avenue, Room 1140; Zoom
NOTE: Ivana Malenica will be presenting virtually.
Abstract
Plug-in estimators based on flexible machine learning methods can exhibit substantial bias. Classical debiasing methods rely on the efficient influence function (EIF), which must be derived and evaluated for each target parameter, creating analytical and computational challenges. We propose a unified framework for kernel debiased plug-in estimation (KDPE), which performs debiasing within a reproducing kernel Hilbert space (RKHS) and simultaneously corrects plug-in bias for a broad class of pathwise differentiable parameters. The KDPE constructs the debiased estimator through iterative regularized likelihood updates, yielding a computationally tractable procedure that avoids explicit influence-function calculations while achieving semiparametric efficiency. We then introduce a refinement based on the universal least favorable submodel, termed ULFS–KDPE. This method defines a data-adaptive debiasing flow in the RKHS, formulated as a nonlinear ordinary differential equation on probability densities. We establish existence, uniqueness, stability, and finite-time convergence of the empirical score along the induced flow, and show that the resulting estimator is regular, asymptotically linear, and simultaneously efficient for a broad class of target parameters. Together, these results provide a unified, influence-function-free framework for efficient estimation in nonparametric models, combining theoretical guarantees with computational tractability. Simulation studies illustrate the methods and support the theoretical findings.
Speaker Bio
Ivana Malenica is an Assistant Professor in the Department of Biostatistics at the University of North Carolina at Chapel Hill (UNC). Previously, Dr. Malenica was a Wojcicki-Troper Data Science Fellow at the Harvard Data Science Initiative and a Postdoctoral Fellow in the Department of Statistics at Harvard University. She earned her Ph.D. in Biostatistics from the University of California, Berkeley. Before that, she studied mathematics and was a fellow at the Translational Genomics Research Institute. Her methodological research lies at the intersection of causal inference, non- and semi-parametric statistics, high-dimensional inference, efficiency theory, and machine learning. She is particularly interested in challenges arising in longitudinal and structured dependent settings, including adaptive sequential experiments, online learning, reinforcement learning, and broader applications in personalized health.