iiinova — where your digital self keeps evolving

Reposted by William Nourse

Patrick Mineault @patrickmineault.bsky.social · 14/09/2026

Are virtual flies playing Beat Saber? SM64? Doom? No. The new fly connectome is an incredible achievement! Less is going on in the sim videos than it seems. I'm stoked to see people engage with neuroscience though. My breakdown: www.neuroai.science/p/are-flies-...

neuroai.science

Are flies playing Beat Saber?

What a time to be alive!

27021

Reposted by William Nourse

Rune W. Berg @runewberg.bsky.social · 28/07/2026

New paper: Spinal circuitry modeled for locomotion using spatial transcriptomics and cell-type projections. Local excitation, long-range inhibition, and a longitudinal skew—an asymmetric "Mexican hat"—give rhythm, rotational dynamics, and gait, with no pacemakers. www.nature.com/articles/s41...

03112

William Nourse @wnourse.bsky.social · 10/07/2026

(11/11) Is this how the nervous system generates polar sequences? Probably not. Is this architecture easy to use? Absolutely, and much of the behavior can be tuned single-shot with no learning. Stay tuned for other papers in the pipeline, where we use these trajectories for fun things...

010

William Nourse @wnourse.bsky.social · 10/07/2026

(10/11) Lastly, why did we bother going through all of this trouble to make this as a SNN vs a state machine or RNN? Well, we can execute this on neuromorphic hardware! We ran our network at various sizes on SpiNNaker2, and saw 2-3 orders of magnitude speed improvements, with 3-4 orders less energy!

Comparing GPU, single and multi-threaded CPU, and SpiNNaker2 for executing our network. SpiNNaker2 does way, way better

100

William Nourse @wnourse.bsky.social · 10/07/2026

(9/11) This output mapping also allows us to fit our trajectory dynamics to arbitrary polar shapes, using essentially a lookup-table since the modulatory parameters are fairly predictable/consistent.

Two funky polar shapes, rose curve on the right and a harmonically distorted curve on the left.

100

William Nourse @wnourse.bsky.social · 10/07/2026

(8/11) But wait, if we change the firing rate of the WTA neurons, then that also changes the rotational speed! To address this, we add a readout population which receives 1:1 input from the WTA. Using the same modulatory mechanism as the speed control, we can get radial ctrl.

Left: schematic of the network with readout neurons added. Changing the stim rate into the readout neurons changes the radius, either uniformly (center) or spatially unique (right). Network size N=50 for butter-smooth curves

110

William Nourse @wnourse.bsky.social · 10/07/2026

(7/11) Polar trajectories require 3 parts: a phase angle, an angular velocity, and a radius. The active neuron identity controls the angle, modulating the interneurons controls the velocity, how do we control the radius? We use the output firing rate, normalized by the fastest possible rate.

A boring polar trajectory, where each neuron is firing at the same rate so the resulting polar trajectory is a circle. A chunky circle with only 5 samples, but a circle regardless.

110

William Nourse @wnourse.bsky.social · 10/07/2026

(6/11) Using gain-control synapses to stimulate the accessory populations (similar to shunting inhibition, but with an excitatory effect), we can control the rate of propagation with complete spatial independence (i.e the transition from a -> b doesn't affect b -> c)

100

William Nourse @wnourse.bsky.social · 10/07/2026

(5/11) Where things get interesting is when we add accessory populations between the WTA nodes. These populations act as AND gates, allowing activity to propagate sequentially when the accessory pop is stimulated. Changing which pop is excited switches the direction (see the top-level video).

100

William Nourse @wnourse.bsky.social · 10/07/2026

(4/11) Using a WTA allows us to constrain the network such that only one "state" is active at a time, assuming that each node in the WTA corresponds to a desired phase angle. Whenever any neuron is stimulated, it silences ongoing activity in the other nodes. Pretty classic.

110

William Nourse @wnourse.bsky.social · 10/07/2026

(3/11) So, how can we make a spiking neural network (SNN) that generates polar trajectories? We take a classic architecture from #compneuro, the winner-take-all (WTA), and we add two things: interneuron populations between the WTA nodes, and a readout layer from the WTA.

A connectivity matrix describing the network architecture, and a schematic diagram

100

William Nourse @wnourse.bsky.social · 10/07/2026

(2/11) Why do we want to make polar trajectories? A lot of biological neural systems responsible for really cool real-world behaviors have dynamics explainable as polar trajectories. This includes head-direction, reaching, and locomotion (my fave, see an example at www.nature.com/articles/s41...)

nature.com

Movement is governed by rotational neural dynamics in spinal motor networks - Nature

A study presents ensemble recordings of neurons in the lumbar spinal cord indicating that activity in spinal cord circuits for movement follows low-dimensional rotational dynamics, and proposes a theo...

130

William Nourse @wnourse.bsky.social · 10/07/2026

How can we generate controllable polar trajectories on neuromorphic hardware? Check out our new pre-print on arXiv to find out all the details, and read the rest of the thread for an overview (1/11): doi.org/10.48550/arX... #neuroAI #CompNeuro #pre-print #neuroskyence #neuromorphic

182

William Nourse @wnourse.bsky.social · 10/07/2026

Honestly, for me it's separating the generation of data from the plotting of data. So, so, so many paper repos generate all of the figures in one big jupyter notebook which will never work on anyone else's machine.

110

Reposted by William Nourse

jordan @jordan.matelsky.com · 15/06/2026

so I just realized I forgot to #skeetprint my Contactomes paper... (is that what we're calling it?) www.biorxiv.org/content/10.6... @kordinglab.bsky.social @dacolon.bsky.social smithneurallab.weebly.com @jhuapl.bsky.social #academicsky #preprint #neuroscience

a single dendritic spine with a handful of synaptic partners but a HUGE number of contact partners

23512

William Nourse @wnourse.bsky.social · 23/04/2026

It’s always an exciting day to see a neuromorphic system on the cover of Nature! Event-based vision is here to stay, now our community needs to show the advantages that neuromorphic approaches can bring to all the other levels of the stack…

010