Lightnews — Scholar-powered news

Tom George

@tomnotgeorge.bsky.social

220 followers 330 following 31 posts

Neuroscience/ML PhD @UCL • NeuroAI, navigation, hippocampus, ... • Open-source software tools for science (https://github.com/RatInABox-Lab/RatInABox) • Co-organiser of TReND CaMinA summer school 🔎👀 for a postdoc position…

Posts Media Videos Starter Packs

Tom George @tomnotgeorge.bsky.social · 2d

woah, are my retinas working, or is that one character away from #RatInABox 👀...

@dlevenstein.bsky.social is right, could be time for a colab

Tom George @tomnotgeorge.bsky.social · Jan 17

Deadline extended until 31st January!!!

Marcus Ghosh @marcusghosh.bsky.social · Jan 9

Join us in Zambia for the third TReND-CaMinA course: computational neuroscience & machine learning in Africa.

📆Applications open until 15.01.

🧠🧪

trendinafrica.org/trend-camina/

A poster advertising the TREND CAMINA summer school.

Tom George @tomnotgeorge.bsky.social · Dec 27

Hey Antonio, great list! Please could you add me too? I consider myself firmly in this space 😁
scholar.google.com/citations?hl...

thanks!

‪PhD, University College London‬ - ‪‪Cited by 124‬‬ - ‪Machine learning‬ - ‪Theoretical neuroscience‬

scholar.google.com

Tom George @tomnotgeorge.bsky.social · Dec 17

and there's this nice work by @neuralreckoning.bsky.social ! www.nature.com/articles/s41...

Neural heterogeneity promotes robust learning - Nature Communications

The authors show that heterogeneity in spiking neural networks improves accuracy and robustness of prediction for complex information processing tasks, results in optimal parameter distribution simila...

Tom George @tomnotgeorge.bsky.social · Nov 28

Very nice work by @sjshipley.bsky.social on place cells and Alzheimers!

Sarah Shipley @shipleysj.bsky.social · Nov 28

Excited to share that our preprint is out!
doi.org/10.1101/2024...

The structure of place cell reactivations was disordered in AD mice compared to WT! This was predictive of reduced place cell stability and memory performance on a radial-arm maze task.

Here are a few details:

1/5🧵

Disordered Hippocampal Reactivations Predict Spatial Memory Deficits in a Mouse Model of Alzheimer's Disease

Alzheimer's disease (AD) is characterised by progressive memory decline associated with hippocampal degeneration. However, the specific physiological mechanisms underlying hippocampal dysfunction in A...

Tom George @tomnotgeorge.bsky.social · Nov 26

Thats great to hear, reach out if you run into any problems!

Tom George @tomnotgeorge.bsky.social · Nov 25

Great question. Local optima will always be hard to identify. Ofc if you have a reason to believe behaviour really _isn't_ a good initialisation then you shouldn't use it.

You can always / we already track the log-likelihood of held-out spikes. If this increases then things are looking good.

Tom George @tomnotgeorge.bsky.social · Nov 25

the sky is bluer*

Tom George @tomnotgeorge.bsky.social · Nov 25

you were right though....the grass is greener over here ;)

Tom George @tomnotgeorge.bsky.social · Nov 25

At the risk of rambling I'll end the thread here and perhaps do a deeper dive in the future. Give it a read (or better, try it on your data) and let us know your thoughts!

tomge.org/papers/simpl/

21/21

SIMPL: Scalable and hassle-free optimisation of neural representations from behaviour

An efficient technique for optimising tuning curves starting from behaviour by iteratively refitting the tuning curves and redecoding the latent variables.

Tom George @tomnotgeorge.bsky.social · Nov 25

This isn’t cheating, behaviour has always been there for the taking and we should exploit it (many techniques specialise in joint behavioural-neural analysis). If we ignore behaviour SIMPL still works but the latent space isn’t smooth and “identifiable”...certainly something to consider.

20/21

Tom George @tomnotgeorge.bsky.social · Nov 25

Initialising at behaviour is a powerful trick here. In many regions (e.g., but not limited to, hippocampus 👀), a behavioural correlate (position👀) exists which is VERY CLOSE to the true latent. Starting right next to the global maxima help makes optimisation straightforward.

Tom George @tomnotgeorge.bsky.social · Nov 25

These non-local dynamics aren’t a new discovery by any means but this is, in our opinion, the correct and quickest way to find them.

18/21

Tom George @tomnotgeorge.bsky.social · Nov 25

And there’s cool stuff in the optimised latent too. It mostly tracks behaviour (hippocampus is still mostly a cognitive map) but does occasional big jumps as though the animal is contemplating another location in the environment.

17/21

Tom George @tomnotgeorge.bsky.social · Nov 25

Dubious analogy: Using behaviour alone to study neural representations (status quo for hippocampus) is like wearing mittens and trying to a figure out the shape of a delicate statue in the dark. Everything is blurred.

16/21

Tom George @tomnotgeorge.bsky.social · Nov 25

The old paradigm of “just smooth spikes against position” is wrong! Those aren’t tuning curves in a causal sense…they’re just smoothed spikes. These “real” tuning curves (the output of an algorithm like SIMPL) are the ones we should be analysing/theorising about.

15/21

Tom George @tomnotgeorge.bsky.social · Nov 25

It’s quite a sizeable effect. The median place cell has 23% more place fields...the median place field is 34% smaller and has a firing rate 45% higher. It’s hard to overstate this result…

14/21

Tom George @tomnotgeorge.bsky.social · Nov 25

When applied to a similarly large (but now real) hippocampal dataset SIMPL optimises the tuning curves. “Real” place fields, it turns out, are much smaller, sharper, more numerous and more uniformly-distributed than previously thought.

13/21

Tom George @tomnotgeorge.bsky.social · Nov 25

SIMPL outperforms CEBRA — a contemporary, more general-purpose, neural-net-based technique — in terms of performance and compute-time. It’s over 30x faster. It also beats pi-VAE and GPLVM.

12/21

Tom George @tomnotgeorge.bsky.social · Nov 25

Let’s test SIMPL: We make artificial grid cell data and add noise to the position (latent) variable. This noise blurs the grid fields out of recognition. Apply SIMPL and you recover a perfect estimate of the true trajectory and grid fields in a handful of compute-seconds.

11/21

Tom George @tomnotgeorge.bsky.social · Nov 25

I think this gif explains it well. The animal is "thinking" of the green location but located at the yellow. Spikes plotted against green give sharp grid fields but against yellow are blurred.

In the brain this discrepancy will be caused by replay, planning, uncertainty and more.

Tom George @tomnotgeorge.bsky.social · Nov 25

behaviour =/= latent.

This is obvious in non-navigational regions. But for HPC/MEC/etc. it’s definitely often overlooked…behaviour alone explains the spikes SO well (read: grid cells look pretty) it’s common to just stop there. But that leaves some error

9/21

Tom George @tomnotgeorge.bsky.social · Nov 25

In order to know the “true” tuning curves we need to know the “true” latent which passed through those curves to generate spikes. i.e. what was the animal thinking of…not what was the animal doing. This latent, of course, is often close to a behavioural readout such as position

Tom George @tomnotgeorge.bsky.social · Nov 25

So what’s the idea inspiring this? Basically, tuning curves (defined as plotting spikes against behaviour) aren’t the brains “real” tuning curves in any causal sense. But often we analyse and theorise about them as though they are. That's a problem.

7/21

Tom George @tomnotgeorge.bsky.social · Nov 25

SIMPL is also "identifiable", returning not just any tuning curves but specifically THE tuning curves which generated the data (there are some caveats / subtleties here)

6/21