A new paper published on arXiv introduces the concept of Structure-agnostic (SA) models, which are designed to account for the lack of prior knowledge about structural assumptions in data-generating laws. While previous work showed that Double Machine Learning (DML) estimators are minimax under these SA models for certain functionals, this paper demonstrates that these DML estimators are asymptotically inadmissible for two of those functionals. The authors propose alternative second-order estimators, specifically empirical higher-order influence function (HOIF) estimators, which asymptotically dominate the DML estimators under the SA model. AI
IMPACT This research may lead to more robust and efficient statistical estimators in machine learning applications where structural assumptions are unknown.
RANK_REASON This is a research paper published on arXiv detailing theoretical findings in statistical machine learning. [lever_c_demoted from research: ic=1 ai=1.0]
- Balakrishnan
- Double Machine Learning
- Gaussian sequence model
- Hoifa Nara
- Liu
- Structure-agnostic models
- U-statistic
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
