Check out the paper and code: arxiv.org/abs/2507.22813

github.com/AdaptiveMoto...

This approach enhances the reliability of trigger reconstruction, making it capable of distinguishing between clean & trojaned models. 🚀

Congrats to all the authors who did an amazing job! 3/4

By employing a diffusion-based generator guided by the target classifier, #DISTIL iteratively produces candidate triggers that align with the model's internal representations associated with malicious behavior. 2/4

And a big welcome to @mirzious.bsky.social to Bluesky! 💙🦋👏 - please follow him; he’s a rising star in merging trustworthy, robust AI for science (just check out his CV 🔥💪): scholar.google.com/citations?us...

Plus the code is #opensource and a Python package for ease of testing and adding to your fav OOD problem 👏

github.com/AdaptiveMoto...

Demo it in Colab, etc!

Stars ⭐️ appreciated! Always helpful to know when to support a code base 😉🥰🍾

#AROS💍 leverages neural ODEs and Lyapunov stability theory to craft an embedding method to smartly detect OOD samples. Strikingly, we can improve performance on popular adversarial detection benchmarks such as CIFAR10 vs CIFAR100 by over 40% 👏

🔥🚀 we are excited to keep pushing this line of work 💪