🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Diffusion models generalize *really* well: if you give them a million pictures of cats, they'll learn to generate reasonable-looking cats no one's ever seen before. But the weird thing is that no one knows why they work! In a theory paper accepted to #ICLR2025, I dug into this.

Hat down to these brave and beautiful hearts. They put us all to shame.

Our latest work on the longstanding problem of learning a (normalized) energy model from data…

Great work by Changde Du from Huiguang He's lab at the Chinese Academy of Sciences. How similar are visual and conceptual representations in (multimodal) large language models to those found in humans? It turns out quite similar! nature.com/articles/s4225…

Two Assistant Professor positions are open at Aalto, respectively in "Neuroscience and Biomedical Engineering" and "Living State Systems", in a department which I can warmly recommend. Feel free to reach out to me for questions. aalto.fi/en/open-positi… aalto.fi/en/open-positi…

Preprint of today: Beyer et al., "Highly Compressed Tokenizer Can Generate Without Training" -- github.com/lukaslaobeyer/… The latent space of tokenizers already provides a good enough abstraction to work with -- you don't have to use a diffusion model on top to inpaint, etc!

As early as primary visual cortex, V1, neural activities are associated with target boosting and distractor suppression in guided visual search, reports this interesting paper Duecker et al 2025, shorturl.at/jcVN5

"Hidden in plain sight: VLMs overlook their visual representations" by Stephanie Fu, Tyler Bonnen, Devin Guillory, Trevor Darrell arxiv.org/abs/2506.08008

I use FSD for over 2 hours of travel every day around the Philadelphia area and have never experienced an issue as bad as this one. Any ideas why it chose the completely wrong lane? Has anyone ever experienced anything like this?

A good language model should say “I don’t know” by reasoning about the limits of its knowledge. Our new work AbstentionBench carefully measures this overlooked skill in leading models in an open-codebase others can build on! We find frontier reasoning degrades models’ ability to

The Kempner Institute's Frontiers in NeuroAI symposium showcased researchers from around the world who are pushing forward the new field of #NeuroAI. Learn about the event & find links to recordings here: bit.ly/3G5Urhv #NeuroAI2025 #AI Sham Kakade Bernardo Sabatini

How is high-level visual cortex organized? In a new preprint with Martin Hebart & Katharina Dobs, we show that category-selective areas encode a rich, multidimensional feature space 🌈 biorxiv.org/content/10.110… 🧵 1/n

Predictive coding in the human olfactory system Opinion by Sam Lyons & Jay Gottfried tinyurl.com/nhznb38v

Diffusion models create novel images, but they can also memorize samples from the training set. How do they blend stored features to synthesize novel patterns?  Our new work shows that diffusion models behave like Dense Associative Memory: in the low training data regime (number

Proud to announce that my stealth team has made a breakthrough that will unlock the next leap: artificial HYPER intelligence, which will make ASI look like a weak and stupid baby. The $1tn seed round for our company, Safe Hyper Intelligent Technologies, is already oversubscribed.

When and why are neural network solutions connected by low-loss paths? In our #ICML2025 paper, we show that mode connectivity often arises from symmetries—transformations of parameters that leave the network’s output unchanged. Paper: arxiv.org/abs/2505.23681 (1/6)

1/2) Happy to share the preprint of our workshop paper on using information theory to find class separation in diffusion models It generalizes previous models of speciation and symmetry breaking to generic class definitions