Olivier Marre

Please RT: postdoc positions open in the lab to work on a @ERC_Research -funded project on retinal circuits and computations. Don't hesitate to DM or email me for details. oliviermarre.free.fr/PostdocPositio…

In this short piece we make the case for latent equivariant operators methods✨, an alternative to classical and equivariant nets that shows promise for out-of-distrib classif. We lay out the challenges ahead for scaling these methods to larger datasets 🧐 follow @minhinhtrng 👀

Modern vision models lacks robustness when objects appear in unusual poses. @StphTphsn1 and I study latent equivariant operators as a remedy and discuss caveats of these operators. Below is a summary of the work, accepted at the GRaM Workshop at ICLR @iclr_conf 2026. 🧵

Focus on the black cross and the red dots will appear to follow waves even though they move straight.

@_sholtodouglas @iskander @animesh_garg @UNC Hi, late to the party but we may have the same problem in my academic bio lab, happy to hear about the options here (also, if you see @DarioAmodei please say Hi from me, we worked together during his PhD 🙂 ).

@iskander @animesh_garg @UNC You should have a dm from him

If any ants follow me: is there any kind of "Claude for an academic bio lab" deal? I want to buy Max subscriptions for everyone in the lab, but can't tell if there's anything simpler than buying them all individually (neither whole-university Academia or Team plans seem right)

Enjoy the read !

A citation from 1744 (no typo !).

How does our visual system process natural scenes ? How can we approach this question ? Happy to share this recent review written with Samuele Virgili where we ask these questions at the level of the retina. sciencedirect.com/science/articl…

Why do video models handle motion so poorly? It might be lack of motion equivariance. Very excited to introduce: Flow Equivariant RNNs (FERNNs), the first sequence models to respect symmetries over time. Paper: arxiv.org/abs/2507.14793 Blog: kempnerinstitute.harvard.edu/research/deepe… 1/🧵

Our @NeurIPSConf 2025 paper - Spotlight selection (~top 3%) - will be presented tomorrow in San Diego by my co–first author @simoneazeglio. We welcome your feedback! 📍 Fri Dec 5, 11 AM-2 PM PST, Exhibit Hall C,D,E #2005 📄 𝐏𝐚𝐩𝐞𝐫: tinyurl.com/ilocal-paper

🧵 What if two images have the same local parts but represent different global shapes purely through part arrangement? Humans can spot the difference instantly! The question is can vision models do the same? 1/15

Our approach: introduce 4 core axioms any meaningful decomposition should satisfy: ✓ Completeness: recovers total mutual information ✓ Locality: local changes have local effects ✓ Positivity: information ≥ 0 ✓ Additivity: combines measurements properly

🧠🔬 Excited to share our #NeurIPS2025 paper: "Convolution Goes Higher-Order"! We asked: Can shallow networks be as expressive as deep ones? Inspired by biological vision, we introduce higher-order convolutions that capture complex image patterns standard CNNs miss. 🧵👇

🧠 How do neurons encode information? We know HOW MUCH, but what about WHAT information they encode? Our new work uses diffusion models to decompose neural information down to individual stimuli & features! 🎯Spotlight at #NeurIPS2025 🌟📄 arxiv.org/abs/2505.11309

@StphTphsn1 @TrueAIHound Ahah yes, this makes the retina a great place to be !

@TrueAIHound very nicely put. In bonus, and to paraphrase my mentor, working in the retina is neat because nobody thinks it is the seat of consciousness @RetinaGeek

Neuroscience bits from my research. I'm neither a fake-AGI accelerationist nor a weirdo doomer. Still, I believe that a breakthrough can happen at any time. I base my argument on neuroscience research. Here it is. Knowing that the retina is perfectly designed for the visual cortex and vice versa, and realizing that it's almost impossible, given the daunting complexity of the task, to unlock the hidden principles of the visual cortex, it makes sense to focus on the design of the retina. The architecture and operation of the retina is very well understood. I believe that trying to understand exactly why it operates the way it does is the best way to flesh out most of the key principles of visual perception. Things like generalization, recognition, on-the-fly representations, prediction, etc. all depend on the kind of signals arriving from the retina. Once visual perception is solved, solving the rest of the intelligence puzzle (motivation, motor control/learning, language module, etc.) will be a piece of cake in comparison. I have strong reasons to believe that all the sensory cortices, including the auditory and tactile cortices, work exactly like the visual cortex. Even olfactory and gustatory work the same way. Who knows? At one point in the future, we may want to build robot chefs that can actually "smell" and "taste" the food. So yes. I believe that a breakthrough can happen at any time. 😮