Merging LoRA weights? You might be throwing away a powerful security feature! 🤯
A new paper shows unmerged LoRA modules are a game-changer for detecting near-OOD data.
I wrote a blog post breaking down the idea. Check it out!
🔗: www.mostafaelaraby.com/paper%20revi...

PhD students: How much time does your supervisor spend with you?

A Nature survey found ~half spend <1 hour/week with theirs.

Yet, more supervisor time is linked to much higher satisfaction (82% vs 69%). Supportive mentorship is crucial.

Full article: www.nature.com/articles/d41...
#PhDLife

Go beyond the data! 🧠 Neural network weights are a rich, untapped resource.
My new blog post dives into Weight Space Learning, exploring how we can analyze models to predict performance and combine them to build better ones.
#ML #DNN #Weight_Space
mostafaelaraby.com/paper%20revi...

GROOD is a practical, post-hoc framework for robust OOD detection & safer AI.

Huge thanks to the team: Mostafa ElAraby, Sabyasachi Sahoo, Yann Pequignot, Paul Novello, Liam Paull.

And collaborators: @Mila_Quebec, @DIRO_UdeM, @UniversiteLaval, & the DEEL project.

The results speak for themselves! 🏆

GROOD achieves SOTA performance on major benchmarks. Highlights:

🔹 84.44% Far-OOD AUROC on CIFAR-100

🔹 94.8% Far-OOD AUROC on ImageNet-1K

🔹 Robust on Transformers (ViT & Swin-T) where others fail!

A picture is worth a thousand words. 📸

While the standard feature space (left) is messy, our gradient-aware space (right) creates a much clearer separation between In-Distribution (blue) and OOD (red/orange) samples!

#DataScience #ComputerVision

What's the secret sauce? ✨

Instead of just feature distance, GROOD uses a synthetic OOD prototype to measure a sample's gradient sensitivity.
ID samples are stable, while OOD samples are sensitive. This creates a powerful, clear signal for detection!

🚀 We're excited to introduce GROOD! A new, training-free method to help models know what they don't know (aka OOD Deteciton).

🔗 Paper: openreview.net/forum?id=2V7...
💻 Code: github.com/mostafaelara...
🎥 Video Overview: www.youtube.com/watch?v=n4-g...

Just dropped a new blog post! 🚀 I'm breaking down how to move your machine learning projects from Docker to Singularity for use in high-performance computing (HPC) environments. Check it out! #MLOps #Docker #Singularity #HPC #MachineLearning
www.mostafaelaraby.com/tutorials/20...

Ever wonder why LLMs sometimes... make things up? 🤔 I dove into a deep survey on AI hallucinations & summarized the key findings! 💡
Check out the breakdown:
www.mostafaelaraby.com/paper%20revi...
#AI #LLM #Hallucinations #NLP #Tech #Research

Topology of Out-of-Distribution Examples in Deep Neural Networks a paper review
open.substack.com/pub/mostafae...

Out-of-Distribution Detection Ensuring AI Robustness
open.substack.com/pub/mostafae...

#OutofDistribution #AIRobustness #DeepLearning #MachineLearning