Mitya

First skeptical, then excited hopefully

Biological Brains Inspire a New Building Block for Artificial Neural Networks Biological Brains Inspire a New Building Block for Artificial Neural Networks on Simons Foundation

#FlatironCCN researchers drew on lessons from neurobiology to enhance artificial systems using a new type of computational component that is more akin to real brains: https://www.simonsfoundation.org/biological-brains-inspire-a-new-building-block-for-artificial-neural-networks/ #science

Sorry, we don't have open positions for post-bachelors.

ReSU: A novel biologically inspired computational primitive for dynamical data. Now at AAAI board 489

Schematic of the early olfactory circuit with local interneurons inhibiting projection neurons. Above local neurons, a simplified depiction of a curved background manifold with a new odor that has an orthogonal component to the manifold. The circuit receives a mixture of background and new odor (apple) and outputs the new odor only.

Why do we stop smelling odors that linger? In our new @prxlife.bsky.social paper with @pfrancois.bsky.social, @gautamreddy.bsky.social, and Massimo Vergassola, we develop a manifold learning theory of this olfactory habituation process in fluctuating environments.
🔗 doi.org/10.1103/q62z...
🧵🔽

Careers Careers on Simons Foundation

Joint junior faculty position in Computational Neuroscience, between Ctr for Computational Neuroscience at @flatironinstitute.org and the CUNY Graduate Center @thegraduatecenter.bsky.social . Application deadline: 16 Jan 2026!

www.simonsfoundation.org/flatiron/car...
cuny.jobs/new-york-ny/...

@hritz.bsky.social Thanks! De Cock & De Moor is a classic—along with Katayama, whom I cite. I wish I could cite everyone in the subspace🙂 I need to read the Larimore paper

A Network of Biologically Inspired Rectified Spectral Units (ReSUs) Learns Hierarchical Features Without Error Backpropagation We introduce a biologically inspired, multilayer neural architecture composed of Rectified Spectral Units (ReSUs). Each ReSU projects a recent window of its input history onto a canonical direction ob...

Move over ReLU 🚀
Meet **ReSU** (Rectified Spectral Unit): a biologically inspired, self-supervised unit for learning from dynamical data. A backprop-free multilayer ReSU network learns predictive features and recapitulates *Drosophila* vision.
To appear at AAAI: arxiv.org/abs/2512.23146

A key point of our work is precisely that a neuron’s meaning comes from the dynamics of the upstream neuronal population it organizes. By learning filters that group inputs with a shared past or future, a neuron clusters dynamical trajectories of the collective dynamics. Where is the fallacy here?

How Sensory Processing Works: Neurons May Predict the Future and Remember the Past - Daily Neuron New research on sensory processing suggests neurons are self-supervised learners that group stimuli by common pasts or futures to make sense of the world.

Here is a neuronal level analysis: dailyneuron.com/how-sensory-...

YouTube video by Brown University CCBS Seminar: “What does the neuron do? A self-supervised dynamical model for neuroscience and AI"

Delighted to talk at Brown about our self-supervised neuronal algorithm for modeling biological circuits—and challenging backprop along the way. Thanks to @leokoz8 for the kind invitation! youtu.be/AF3Uhrm__U4?...

Biological neurons cluster dynamical stimulus trajectories to predict what’s coming and infer what just happened. If you’re at NeurIPS, stop by our poster #2107 — on display now until 2pm

Interested in algorithmic neuron models and learning rules at the intersection of neuroscience, AI, dynamical systems, and control theory? Working towards a PhD in neuroscience, physics, EE, or math? Apply for a summer internship with us: apply.interfolio.com/177775 At NeurIPS? Feel free to DM me.

Interested in algorithmic neuron models and learning rules at the intersection of neuroscience, AI, dynamical systems, and control theory?
Have a PhD in neuroscience, physics, EE, or math? Consider joining us: apply.interfolio.com/173400
If you’re at NeurIPS, feel free to DM me.

An exciting opportunity!

Thanks @hadivafaii.bsky.social for the invitation! x.com/hadivafaii/s...

How Sensory Processing Works: Neurons May Predict the Future and Remember the Past - Daily Neuron New research on sensory processing suggests neurons are self-supervised learners that group stimuli by common pasts or futures to make sense of the world.

The brain survives by predicting the future. We suggest that single neurons cluster trajectories with common futures. But when observations are noisy, retrospection helps. Even individual neurons may look backward—like LGN lagged cells and olfactory bulb mitral cells: dailyneuron.com/how-sensory-...

Our poster in San Diego now at Society for Neuroscience

Fantastic and timely workshop! Many thanks to the organizers! neuroai-multimodal-workshop.github.io

Neural Network: Dmitri Chklovskii Spotify video

The neuron as a controller and a few other thoughts. Thanks to the GreyMatters podcast for hosting me: open.spotify.com/episode/1ox9...