Hyperparticle/one-pixel-attack-keras
A Keras implementation demonstrating how to fool deep neural networks by modifying a single pixel using differential evolution.
Not currently ranked — collecting fresh signals.
star history
This project reproduces a research paper on adversarial attacks against CNNs, specifically targeting Cifar10 and ImageNet classifiers. It uses differential evolution, an evolutionary algorithm, to iteratively find single-pixel modifications that cause neural networks to misclassify images. The implementation provides Jupyter notebooks as tutorials and operates as a black-box attack requiring only prediction probabilities.
Frequently asked
- What is Hyperparticle/one-pixel-attack-keras?
- A Keras implementation demonstrating how to fool deep neural networks by modifying a single pixel using differential evolution.
- Is one-pixel-attack-keras open source?
- Yes — Hyperparticle/one-pixel-attack-keras is open source, released under the MIT license.
- What language is one-pixel-attack-keras written in?
- Hyperparticle/one-pixel-attack-keras is primarily written in Jupyter Notebook.
- How popular is one-pixel-attack-keras?
- Hyperparticle/one-pixel-attack-keras has 1.2k stars on GitHub.
- Where can I find one-pixel-attack-keras?
- Hyperparticle/one-pixel-attack-keras is on GitHub at https://github.com/Hyperparticle/one-pixel-attack-keras.