SabaPivot's picture
|
download
raw
1.37 kB
# Source manifest
- Primary paper: https://arxiv.org/abs/2602.22130
- Versioned PDF: https://arxiv.org/pdf/2602.22130v1
- TeX source archive used to transcribe Algorithm 1 and Theorems 3.2/4.1: https://export.arxiv.org/e-print/2602.22130
- OpenReview submission: https://openreview.net/forum?id=no9dQDBxsu
- Author-hosted PDF: https://cseweb.ucsd.edu/~dakane/MeanShiftSampleComplexity.pdf
- Code search: no official implementation or code URL appears in the paper, OpenReview record, arXiv metadata, or title/author GitHub searches performed 2026-07-21.
- Independent implementations: `reproduce_full.py` implements the Algorithm 1 ECF ratio and minimax tournament, literal finite covers in dimensions 1 and 2, a documented high-dimensional continuous diagnostic, and LP-constructed lower-bound pairs. `exact_d3_certification.py` extends the literal Cartesian-cover tournament to dimension 3, certifies the lattice cover radii analytically, scores every candidate against every active frequency, and runs both normalized-budget and full-Theorem-3.2-expression interventions. The d=4–16 continuous optimizer is a non-equivalent diagnostic, not literal Algorithm 1. No external dataset, model, or HF Job is used.
The earlier `reproduce.py` is retained only as provenance for the superseded low-score run. It is not the evidentiary implementation for the 2026-07-21 rerun.

Xet Storage Details

Size:
1.37 kB
·
Xet hash:
fd45196f7d5f466407d5633fabbafbf417a3c0a7202b19162acff84f65a2cdb3

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.