Gabriele Venturato

We developed a library to make logical reasoning embarrasingly parallel on the GPU.

For those at ICLR 🇸🇬: you can get the juicy details tomorrow (poster #414 at 15:00). Hope to see you there!

#ai #llms #reasoning #deepseek #3sat #iclr | Rishi Hazra 🚨 𝗡𝗲𝘄 𝗣𝗮𝗽𝗲𝗿 𝗔𝗹𝗲𝗿𝘁: 𝗛𝗮𝘃𝗲 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 𝗟𝗲𝗮𝗿𝗻𝗲𝗱 𝘁𝗼 𝗥𝗲𝗮𝘀𝗼𝗻? Despite their impressive performance on reasoning tasks, LLMs often succeed by fitting to patterns rather than understanding logic. Are...

What is reasoning and have LLMs learned it?

We answer these two questions in our new pre-print using the phase transition in random 3-SAT.
arxiv.org/abs/2504.039...

Together with @rishihazra95.bsky.social, @gabventurato.bsky.social and @lucderaedt.bsky.social

Here is also a longer thread:

If you're at #AAAI2025, come check out our demo on neurosymbolic reinforcement learning with probabilistic logic shields 🤖 Tomorrow (Sat, March 1) from 12:30–2:30 PM during the poster session 💻

We all know backpropagation can calculate gradients, but it can do much more than that!

Come to my #AAAI2025 oral tomorrow (11:45, Room 119B) to learn more.

If you're at #AAAI2025, come check out my talk tomorrow (Fri, Feb 28) from 9:30–10:45 AM in the Machine Learning track! 👨🏻‍🏫

📍 Swing by our poster later in the day for more discussion!

🚀 Applications include:
- Safe reinforcement learning
- Logical control in language models
- Temporal reasoning with strong guarantees

Let's discuss! Thoughts, questions, or collaborations? ⬇️ #neurosymbolic #AAAI2025

🙌 @lennertds.bsky.social, @giuseppemarra.bsky.social, @lucderaedt.bsky.social

🔬 In experiments, NeSy-MMs outperform existing neurosymbolic & deep models on both generative & discriminative tasks.

They generalize better to out-of-distribution settings and allow test-time constraint adaptation, making them robust & versatile.

NeSy-MMs integrate:
✅ Relational probabilistic reasoning
✅ Logical constraints
✅ Neural learning
✅ Approximate Bayesian inference

We introduce a novel differentiable particle filter that enables efficient inference & learning while maintaining logical consistency.

Markov models are great for sequential problems, but deep models struggle with constraints, and neurosymbolic AI doesn’t scale.

NeSy-MMs bridge this gap: a new class of differentiable models that provably satisfy relational logical constraints while scaling efficiently.

🔥 Can AI reason over time while following logical rules in relational domains? We will present Relational Neurosymbolic Markov Models (NeSy-MMs) next week at #AAAI2025! 🎉

📜 Paper: arxiv.org/pdf/2412.13023
💻 Code: github.com/ML-KULeuven/...

🧵⬇️

🚀 Do you care about safe AI? Do you want RL agents that are both smart & trustworthy?

At #AAAI2025, we present our demo for neurosymbolic RL—combining deep learning with probabilistic logic shields for safer, interpretable AI in complex environments. 🏰🔥
🧵👇
(1/8)

🦕The 19th conference on Neurosymbolic AI will be in beautiful Santa Cruz (CA, USA), September 8-10, 2025!

CFP is now out: 2025.nesyconf.org/call-for-pap...
🚨 Paper deadline: Feb 28 (abstract), March 7 (full)

#neurosymbolic #NeSy2025

I made a starter pack with the people doing something related to Neurosymbolic AI that I could find.

Let me know if I missed you!
go.bsky.app/RMJ8q3i

Meet our researchers from the DTAI lab at KU Leuven!

Using this starter pack, you can keep up with all the AI research from our PhD students, post-docs, professors and alumni 🦋