marcusghosh - Network

I recently gave a talk on: Nonlinearity and network topology in multimodal circuits. Let me know any...

I recently gave a talk on:

Nonlinearity and network topology in multimodal circuits.

Let me know any thoughts:
https://www.youtube.com/watch?v=duzTIn63WTs&ab_channel=ICNS

Had a great week setting up at both Imperial I-X and EEE.Let me know if you're in London and want to discuss multimodal integration or...

Had a great week setting up at both Imperial I-X and EEE.

Let me know if you're in London and want to discuss multimodal integration or #NeuroAI more broadly!

#introduction update I'm a postdoc based in London (at Imperial I-X).I use neural network models and experimental data to understand how...

#introduction update

I'm a postdoc based in London (at Imperial I-X).

I use neural network models and experimental data to understand how brains combine information across multiple senses (multimodal integration) #neuroscience .

I work with Dan Goodman (@neuralreckoning).

Read our recent work here:
https://doi.org/10.1101/2023.07.24.550311

Join us to hear @briandepasquale discuss his recent paper & more! When: May 3rd (15:00 CEST)Where:...

Join us to hear @briandepasquale discuss his recent paper & more!

When: May 3rd (15:00 CEST)
Where: https://www.world-wide.org/Neuro/SNUFA/

Join us to learn about Crescent Loom A #neuroscience "game" in which you create a creature and explore how neural circuits...

Join us to learn about Crescent Loom

A #neuroscience "game" in which you create a creature and explore how neural circuits generate movement!

When: April 5th (17:00 CET)
Where: https://www.world-wide.org/Neuro/SNUFA/

Learn more: https://crescentloom.com/

Join us to learn about spatially-embedded RNNs from @achterbrain Feb 1st 15:00 CETDetails: https://www.world-wide.org/Neuro/SNUFA/

Join us to learn about spatially-embedded RNNs from @achterbrain

Feb 1st
15:00 CET

Details:
https://www.world-wide.org/Neuro/SNUFA/

Data aside, I was blown away by the level of detail in the paper and how ambitious mapping a larger connectome seems. As Larry Abbott etc...

Data aside, I was blown away by the level of detail in the paper and how ambitious mapping a larger connectome seems.

As Larry Abbott etc point out (below) going from a worm to a mouse = a 10-million-fold increase in brain volume.

https://www.sciencedirect.com/science/article/pii/S0092867420310011

🧵​ 6/6

Zigzag motifs! There are brain-nerve cord zigzag motifs between descending (DN) and ascending neurons (AN). E.g., DN -> AN -> DN1...

Zigzag motifs!

There are brain-nerve cord zigzag motifs between descending (DN) and ascending neurons (AN).

E.g., DN -> AN -> DN1 etc.

Intuitively, these could control sequences of actions!

🧵​ 5/6

There is lots of recurrence! 41% of neurons were involved in polysynaptic recurrent loops (e.g., neuron 1 –> 2 –> 3 –> 1). ...

There is lots of recurrence!

41% of neurons were involved in polysynaptic recurrent loops (e.g., neuron 1 –> 2 –> 3 –> 1).

🧵​ 4/6

Almost everything is multimodal! The maggot has 12 types of sensory neurons (visual, thermal, olfactory etc), but beyond these only 12% of...

Almost everything is multimodal!

The maggot has 12 types of sensory neurons (visual, thermal, olfactory etc), but beyond these only 12% of neurons are unimodal (receive input from a single sense), the other 88% are multimodal!

Plus, most multimodal neurons (62%) integrate information from all modalities!

While, it seems intuitive that deeper layers in a neural network would be progressively more multimodal, this seems quite different to (many) block diagram models (e.g., visual pathways).

🧵​ 3/6

Only 66% of connections are between axons and dendrites! So, the neuron-neuron connectome is actually 4 (or more?) matrices (axon-dendrite,...

Only 66% of connections are between axons and dendrites!

So, the neuron-neuron connectome is actually 4 (or more?) matrices (axon-dendrite, axon-axon, dendirte-dendite and dendite-axon).

Also, these “differentially contribute to feedforward and feedback pathways”.

WILD.

🧵​ 2/6

Lessons from a maggot. I just finished reading this awesome paper from @albertcardona and colleagues, in which they map the Drosophila...

Lessons from a maggot.

I just finished reading this awesome paper from @albertcardona and colleagues, in which they map the Drosophila larva connectome, which is made up of 3,013 neurons and 544,000 synapses.

https://www.biorxiv.org/content/10.1101/2022.11.28.516756v1

The paper is full of interesting results, but here are my 4 favorites!

🧵​ 1/6

Really enjoyed this review of NeuroAI in 2022 from @patrickmineault Especially as he (correctly) places spiking neural networks at the heart...

Really enjoyed this review of NeuroAI in 2022 from @patrickmineault

Especially as he (correctly) places spiking neural networks at the heart of NeuroAI 🧡​

https://xcorr.net/2023/01/01/2022-in-review-neuroai-comes-of-age/

For the first #SNUFA seminar of the year, we'll hear from @lowrank_adrian January 11th15:00 CET Details here:...

For the first #SNUFA seminar of the year, we'll hear from @lowrank_adrian

January 11th
15:00 CET

Details here:
https://www.world-wide.org/Neuro/SNUFA/

Last Friday there was a conference to consider Horace Barlow's work and its ongoing impact. Of all of his papers, most speakers...

Last Friday there was a conference to consider Horace Barlow's work and its ongoing impact.

Of all of his papers, most speakers focussed on two (below).

Strikingly, these two papers contain no models or experimental data, but instead present 💡ideas💡​ about how neurons might process information.

For me, this focus underscores that what we need in #neuroscience are not larger neural networks or larger-scale recordings, but new ideas and ways of thinking about the brain.

Papers:
🧠​ Possible principles underlying the transformations of sensory messages (Barlow 1961)

🧠 Single units and sensation: A neuron doctrine for perceptual psychology? (Barlow 1972)

#introduction I'm a #postdoc based between Paris and London. I use neural network models and experimental data to understand how brains...

#introduction

I'm a #postdoc based between Paris and London.

I use neural network models and experimental data to understand how brains combine information across multiple senses (multimodal integration) #neuroscience .

I work with Dan Goodman (@neuralreckoning) and Volker Bormuth.

I help to run both:
🧠 COMOB - an open-source effort to model the brain: https://comob-project.github.io/

🐩 SNUFA - a monthly #neuro seminar series: https://www.world-wide.org/Neuro/SNUFA/