Pinned post

The latest piece of my PhD work is now published! Check it out at nature.com/articles/s41586-024

We explain how correlated responses in the retinal output may arise when nonlinear receptive fields are stimulated with natural scenes. We think that these concerted responses violate the decorrelation prediction of efficient coding in a cell-type-specific manner in both marmosets and mice.

Through monitoring a large group of marmosets, researchers have categorised their behaviours to find baselines for typical and atypical responses that can help us study psychiatric disorders.
elifesciences.org/reviewed-pre

"[It’s] the largest unexplained number in brain science. I feel like neuroscience should pay more attention to it ..." said Markus Meister when Claudia López Lloreda sat down with him and Jieyu Zheng for a Q&A about the brain's information-processing rate.

thetransmitter.org/computation

Say you take all 12 eggs out of an egg carton, randomly permute them, and put them back in. How many eggs come back to their original place, on average?

One!

What's the probability that exactly one gets back to its original place? This is a lot harder, because unlike the first question it really depends on the number "12". But the answer is close to 1/e. And if we did it for 100 eggs, or a million, the answer would get even closer to 1/e.

What's the probability that exactly n eggs each get back to their original place? Now things get really interesting. The answer is complicated, but again it simplifies a lot in the limit where we permute a huge number of eggs. Then the answer approaches 1/e divided by n factorial.

What's interesting about that? It's the same as the answer to *this* question: if you're standing in the rain, and on average one raindrop lands on your head every second, what's the probability that in one second exactly n raindrops land on your head? At least this is true if raindrops are falling randomly in the most reasonable way - a so-called 'Poisson distribution'.

So random permutations are connected to Poisson distributions. And the connection goes a lot further than I've explained so far.

I've been trying to understand this better and better. The formulas I'm trying to understand are already known, but I've been proving them using category theory:

golem.ph.utexas.edu/category/2

This gives a deeper outlook: instead of just proving an equation about probabilities, we can show that two categories are equivalent, and this has the equation as an easy spinoff. All the fun facts I just listed, and more, become facts about categories!

I never heard of these whole ass deeply evolutionarily preserved organelles in my cell bio class. Anyone know what these dang things are?
en.m.wikipedia.org/wiki/Vault_

We are pleased to announce the 2025 Paris Spring School in Optical Imaging and Electrophysiological Recording in Neuroscience. AKA the Paris Neuro Course.

The course will run 14-27 May 2025.

parisneuro.ovh/

Please boost (with apologies for duplicates).

I wonder if the tools of digital stylometry can detect what people seem to sense when presented with AI-generated text. I don't mean using deep learning to have built up an internalised semantic picture of what type of text LLMs generate, but just purely examining style with eg the count of function words, the use of starting sentences with linking adverbs ("However,..."), having paragraphs all about the same length made of sentences of fairly homogeneous style and so on. This type of analysis doesn't return an informative yes/no, but all kinds of comparative statistics, which people can then iterate on, or drill into, and make judgements about—that is, the software doesn't tell you if a piece of text is AI-generated, it merely unpacks a description of the text that isn't obvious on surface reading. Maybe all this happens under the hood on tools that use ML to classify text, but I much prefer seeing the actual data, not the opaque "answer".

Neuroscience textbooks can be prohibitively expensive for some undergraduate students. A new open-access alternative seeks to change that. Francisco J. Rivera Rosario sat down for a Q&A with Liz Kirby to discuss open-access neuroscience in the classroom.

thetransmitter.org/books/open-

Looks like I will have some funding to hire a computational tech for ~6 months, starting super duper soon! Can be done remotely. Looking for good computational & programming skills with a strong interest in or knowledge of systems neuroscience/hippocampus. PLS SHARE + RT!

I was interviewed on the podcast The One You Feed with Eric Zimmer. We discuss #morality, #addiction, and the general structure of #DecisionMaking.

check it out!

oneyoufeed.net/changing-how-we

Bryan W. Jones @BWJones spoke with The Transmitter's Angie Voyles Askham @avaskham about how the beauty of the retina drew him into vision research and why photography reminds him of the value of that work.

thetransmitter.org/craft-and-c

🥳 BIG NEWS for the lab! 🥳 I'm excited to join the Göttingen neuroscience community and the excellence cluster
@MBExC_de as a tenure-track group leader at the European Neuroscience Institute (ENI-G), where we will continue neural circuit research in a great research environment!
mbexc.de/the-mbexc-welcomes-ju

Every time I heard that the vertebrate retina does not receive any feedback from the central brain I expressed scepticism, and was told repeatedly that no, feedback fibers running backwards along the optic nerve were never found.

Now*, Sylvia Schröder et al. reports that:

"Arousal modulates retinal output", Schröder et al. 2020 cell.com/neuron/fulltext/S0896

Recordings with neuropixels probes on the optic tract of the mouse showed changes in retinal ganglion cell (RGC) activity (the output neurons of the retina whose axons make up the optical tract) in concordance with the arousal state of the mouse. They also measured activity of RGC axon boutons in the superior colliculus.

What a fantastic piece of work.

* for long values of "now".

#neuroscience #neuropixels #vision #retina #mouse

Had never received such a high number of summer student internship applications that read more or less the same: same tone, same paragraph ordering and size, same highlights of my research, same interests. And some critical words in quotes.

I can't think of any explanation other than ChatGPT. What a blight.

#academia

Didier Raoult now published a 'preprint' claiming I am 'close' to Gates Foundation - citing a France Soir article as the reference. He also suggests that my poor facial recognition skills are linked to autism, with 'serious psychosocial consequences'. What a sad man. hal.science/hal-04795904/

Don't say #science or #academia are useless, I just spent three hours showing magnetic or #astronomy wonders to about 250 children in two events, organised by the University of Bologna, so that their parents could enjoy wild shopping or aperitives in the city centre for some time, and it's a service to society I was happy to provide!
The kids seemed happy too.
#astrodon

It's taken 22 years, but public transport and archaeology meet in Thessaloniki this weekend with the opening of its new metro system, which will reduce traffic congestion by as many as 60,000 vehicles.
There are 13 'archaeostations' where ancient finds are displayed or are part of the infrastructure.

#Transport #Archaeology #Thessaloniki #Greece #PublicTransit

theguardian.com/world/2024/nov

Persistence of spike protein at the skull-meninges-brain axis may contribute to the neurological sequelae of COVID-19

Some unrest and abuse about this paper- Parts of the pre-print have been abused previously. What does it show?

cell.com/cell-host-microbe/ful
1/10

Predicting the results of experiments is challenging, especially in fields like neuroscience. This recent study shows that fine-tuned LLMs, like BrainGPT, can predict neuroscience results more accurately than human experts on benchmarks. 🧪

nature.com/articles/s41562-024

Show older
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.