Paint: Accurate, efficient, and sparse Model Inversion via peripheral sparsification and anchor alignment
PLOS One · 17.11.2026 · 21 (11) · e0362597
DOI: 10.1371/journal.pone.0362597 ↗Kurz gefasst
Fachartikel in der Zeitschrift PLOS One, online erschienen am 17.11.2026. Verfasst von Jaehyeon Choi und 2 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Institute for Information and Communications Technology Promotion, Institute of Engineering Research, Seoul National University, Institute of Computer Technology at Seoul National University.
Abstract (Originalsprache)
How can we reconstruct training data from a pre-trained model without access to the original dataset? Model Inversion (MI) attempts to synthesize samples similar to the training dataset. This capability is particularly useful in data-free applications, such as data-free quantization and data-free knowledge transfer, where synthetic data replace inaccessible training samples. For these applications, MI needs to generate high-fidelity samples to preserve semantic information while remaining computationally efficient, enabling scalable generation for reliable downstream performance. However, existing MI methods fail to achieve both high fidelity and efficiency simultaneously. In this paper, we propose Paint ( P eripheral sparsification and A nchor-aligned IN version T echnique), an accurate, efficient, and sparse MI framework for generating high-fidelity synthetic data. Paint integrates low-pass filtering during inversion to suppress high-frequency noise, aligns salient regions toward anchor with peripheral sparsification to preserve object structures during sparse inversion, and exploits soft targets to better capture the predictive uncertainty. Extensive experiments demonstrate that Paint improves inversion efficiency for ViTs and achieves accuracy gains of up to 5.33%p in data-free scenarios.
Förderung
- Institute for Information and Communications Technology Promotion
- Institute of Engineering Research, Seoul National University
- Institute of Computer Technology at Seoul National University
Zitieren
APA
Choi, J., Kim, M., & Kang, U. (2026). Paint: Accurate, efficient, and sparse Model Inversion via peripheral sparsification and anchor alignment. PLOS One, 21(11), e0362597. https://doi.org/10.1371/journal.pone.0362597
Vancouver
Choi J, Kim M, Kang U. Paint: Accurate, efficient, and sparse Model Inversion via peripheral sparsification and anchor alignment. PLOS One. 2026;21(11):e0362597. doi:10.1371/journal.pone.0362597
BibTeX
@article{Choi2026,
author = {Choi, Jaehyeon and Kim, Minjun and Kang, U},
title = {Paint: Accurate, efficient, and sparse Model Inversion via peripheral sparsification and anchor alignment},
journal = {PLOS One},
year = {2026},
volume = {21},
number = {11},
pages = {e0362597},
doi = {10.1371/journal.pone.0362597}
}
EUVIMED