Lottery Prior: Randomized Neural Compression for Zero-Shot Inverse Problems

Haotian Wu, Di You, Pier Luigi Dragotti, Deniz Gunduz
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:137390-137426, 2026.

Abstract

We study zero-shot inverse problems, where a clean signal is recovered from a single degraded observation without external training data. Contrary to the common belief that such problems require highly complex models, we show that a lightweight neural network, when combined with entropy and complexity regularization in a compression-based formulation, is sufficient for high-quality restoration. We propose Lottery Prior, a compression-based inverse solver that leverages architectural priors from random networks and induces a family of implicit priors through randomness, enabling ensemble-based refinement. We further derive non-asymptotic error bounds for compression-based maximum-likelihood inverse solvers, revealing how rate–distortion constraints act as implicit regularizers. Experiments on denoising, noisy super-resolution, and inpainting demonstrate that our method achieves state-of-the-art with significantly fewer effective parameters. Project page: https://eedavidwu.github.io/LotteryPrior/

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wu26bu, title = {Lottery Prior: Randomized Neural Compression for Zero-Shot Inverse Problems}, author = {Wu, Haotian and You, Di and Dragotti, Pier Luigi and Gunduz, Deniz}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {137390--137426}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wu26bu/wu26bu.pdf}, url = {https://proceedings.mlr.press/v306/wu26bu.html}, abstract = {We study zero-shot inverse problems, where a clean signal is recovered from a single degraded observation without external training data. Contrary to the common belief that such problems require highly complex models, we show that a lightweight neural network, when combined with entropy and complexity regularization in a compression-based formulation, is sufficient for high-quality restoration. We propose Lottery Prior, a compression-based inverse solver that leverages architectural priors from random networks and induces a family of implicit priors through randomness, enabling ensemble-based refinement. We further derive non-asymptotic error bounds for compression-based maximum-likelihood inverse solvers, revealing how rate–distortion constraints act as implicit regularizers. Experiments on denoising, noisy super-resolution, and inpainting demonstrate that our method achieves state-of-the-art with significantly fewer effective parameters. Project page: https://eedavidwu.github.io/LotteryPrior/} }
Endnote
%0 Conference Paper %T Lottery Prior: Randomized Neural Compression for Zero-Shot Inverse Problems %A Haotian Wu %A Di You %A Pier Luigi Dragotti %A Deniz Gunduz %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wu26bu %I PMLR %P 137390--137426 %U https://proceedings.mlr.press/v306/wu26bu.html %V 306 %X We study zero-shot inverse problems, where a clean signal is recovered from a single degraded observation without external training data. Contrary to the common belief that such problems require highly complex models, we show that a lightweight neural network, when combined with entropy and complexity regularization in a compression-based formulation, is sufficient for high-quality restoration. We propose Lottery Prior, a compression-based inverse solver that leverages architectural priors from random networks and induces a family of implicit priors through randomness, enabling ensemble-based refinement. We further derive non-asymptotic error bounds for compression-based maximum-likelihood inverse solvers, revealing how rate–distortion constraints act as implicit regularizers. Experiments on denoising, noisy super-resolution, and inpainting demonstrate that our method achieves state-of-the-art with significantly fewer effective parameters. Project page: https://eedavidwu.github.io/LotteryPrior/
APA
Wu, H., You, D., Dragotti, P.L. & Gunduz, D.. (2026). Lottery Prior: Randomized Neural Compression for Zero-Shot Inverse Problems. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:137390-137426 Available from https://proceedings.mlr.press/v306/wu26bu.html.

Related Material