license: mit
library_name: pytorch
datasets:
- uoft-cs/cifar10
pipeline_tag: image-classification
tags:
- computer-vision
- cnn
- cifar10
- adversarial-robustness
- downsampling
- anti-aliasing
NullGuard
Overview
this repository contains a PyTorch CNN trained on CIFAR-10 for image classification with the emphasizs of robustness to adversarial pertubation that targets downsampling operations.
print("tired")
Problem
The composite operator maps a high dimensional space to a low-dimensional space.
By the Rank-Nullity thereom, this matrix has a massive NULL space. An attacker can utilize the discarded samples of these downsampling operations as extra degrees of freedom. By sculpting their adversarial input to be partially within the null space, this allows them to spread the pertubration, minimizing the structural disruption to the original image
Citation
If you use this model or repository in your research, please cite:
@misc{MamaPearlModel,
author = {MamaPearl},
title = {},
year = {2026},
publisher = {GitHub},
url = {https://github.com}
}
License
This project is licensed under the MIT License. For further information, please see the LICENSE file.
Authors
- MamaPearl - Initial Work - MamaPearl GitHub