Mastodawn

adversarial-designs May 1, 2024

Generative image models have gotten a lot better over the last two years, to the point where they are usable in things like marketing.

This puts artists in a bit of a weird position, since other people can use these models to avoid hiring them.

A couple of tools have emerged to help artists protect themselves, like Glaze and Nightshade. Using them is not a guaranteed defense, and also engenders legal risk.

Learn about why this is and how these tools work here: https://adversarial-designs.shop/blogs/blog/how-glaze-and-nighshade-try-to-protect-artists

How Glaze and Nighshade try to protect artists

Generative AI models have become increasingly effective at making usable art. Where does this leave artists? They can use tools like Glaze and Nightshade to discourage others from training models to reproduce their art, but this might not always work, and carries legal risks.

adversarial designs

adversarial-designs Aug 16, 2023

Adversarial patches are hard to defend against because they are robust to denoising-based defenses. A better strategy is generating several partially occluded versions of the input image, getting a set of predictions, and then taking the *least common* predicted label.

https://adversarial-designs.shop/blogs/blog/minority-reports-yes-like-the-movie-as-a-machine-learning-defense

Minority reports (yes like the movie) as a machine learning defense

Adversarial patch attacks are hard to defend against because they are robust to denoising-based defenses. A more effective strategy involves generating several partially occluded versions of the input image, getting a set of predictions, and then taking the *least common* predicted label.

adversarial designs

adversarial-designs Aug 1, 2023

If you build and maintain a database of "fingerprints" of adversarial attacks, you can estimate which kind is being used against your model in real time. This tells you both about the technical sophistication of your adversary, and the strength of possible adversarial defenses.

Learn more at https://adversarial-designs.shop/blogs/blog/know-thy-enemy-classifying-attackers-with-adversarial-fingerprinting

#ThreatIntelligence #AdversarialML

Know thy enemy : classifying attackers with adversarial fingerprinting

In threat intelligence, you want to know the characteristics of possible adversaries. In the world of machine learning, this could mean keeping a database of "fingerprints" of known attacks, and using these to inform real time defense strategies if your inference system comes under attack. Would you like to know more?

adversarial designs

adversarial-designs Apr 22, 2023

When you're defending your computer vision models against adversarial attacks, you might just... need more JPEG

https://adversarial-designs.shop/blogs/blog/what-if-adversarial-defenses-just-need-more-jpeg

What if adversarial defenses just need more JPEG?

Adversarial patterns are specially crafted image perturbations that trick models into producing incorrect outputs. Applying JPEG compression to the inputs of a computer vision model can effectively "smear" out adversarial perturbations, making it more difficult to successfully launch an adversarial attack.

adversarial designs

adversarial-designs Apr 1, 2023

This #AprilFoolsDay, you have the power to turn any laptop into an #Apple.

https://adversarial-designs.shop/products/this-sticker-is-an-apple

this sticker is an apple

adversarial-designs Feb 15, 2023

Most uses of adversarial machine learning involve attacking or a model that someone else has designed. However, you can use the same method to change the deployment environment, so that your model performs better.

This paper was originally published at Neurips 2021.

https://adversarial-designs.shop/blogs/blog/anti-adversarial-examples-what-to-do-if-you-want-to-be-seen

Anti-adversarial examples: what to do if you want to be seen?

Most uses of adversarial machine learning involve attacking or bypassing a computer vision system that someone else has designed. However, you can use the same tools to generate "unadversarial" examples, that give machine learning models much better performance when deployed in real life.

adversarial designs

adversarial-designs Jan 17, 2023

Could the person you're chatting with on mastodon be a 🤖? At DEFCON last year, Justin Hutchens showed a proof of concept using LLMs in conjunction with dialogue management software and selenium to conduct massive automated phishing operations.

Why bring this up now? With the release of ChatGPT, automated phishing attacks have become a lot easier. Read more about it here: https://adversarial-designs.shop/blogs/blog/taking-chatgpt-on-a-phishing-expedition

Taking ChatGPT on a phishing expedition

Are you sure the person you're chatting with online is real? Recent progress in language models like ChatGPT have made it shockingly easy to create bots that perform phishing operations on users at scale.

adversarial designs