noise-weighted-subspace-reconstruction v1.0 (MLST submission release): code for "When Does Separable Structure Pay? Restriction Cost and Covariance Calibration in Noise-Weighted Reconstruction"
Year: 2026
Type: software
Abstract: Experiment code for the manuscript "When Does Separable Structure Pay? Restriction Cost and Covariance Calibration in Noise-Weighted Reconstruction" (D. Wong, Karlsruhe Institute of Technology; submitted to Machine Learning: Science and Technology). The release holds the kits of every study the paper reports, with the run records stored in the repository. These include the prospective crossing test (Y19), the baseline kit (ZC), the plug-in study (E2X+), the measured-noise test on HCP empty-room MEG recordings (registered at https://osf.io/n2rcu/) and the post hoc optimizer refit (X-OPT). Freeze identities are the repository tags y19-registration-20261012, zc-registration-20261013, e2xplus-registration-20261023 and xopt-freeze-20261108(-p2, -p3). The paper's derived evidence, claim-to-evidence table, verification scripts and frozen registration packages are in Zenodo record https://doi.org/10.5281/zenodo.24295048. The repository contains no HCP recordings and no per-record HCP-derived arrays; outputs of the HCP run are in the restricted record https://doi.org/10.5281/zenodo.24295070. Licence: MIT. Funding: Alexander von Humboldt Foundation, through the Alexander von Humboldt Professorship of Markus Klute at the Karlsruhe Institute of Technology.
Author: Wong, Dowling
Institution: Karlsruhe Institute of Technology
Language: English
Subfield: Mathematical Physics
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7234329272
