Postdoc in Computational Neuroscience | CRM Barcelona
Mostly interested in the mechanisms underlying learning, forgetting, memory formation, and most recently also creativity. And "representational drift".
https://jb-eppler.github.io/
Jens-Bastian Eppler
Loading...
New preprint:
Random network structure stabilizes neural manifolds
We’re excited to share our new work on representational drift.
doi.org/10.64898/202...
Representational drift poses a puzzle. 👇
A short thread below. In the next days a figure by figure walk through will follow.
1/5 🧪🧠
Wanna do neuroscience in Paris but can't find interesting lab?
Want to come do a sabbatical but don't know who to collaborate?
Check this webpage aggregating ~all the neuroscience labs (+200) in Paris.
⚠️only the information of 'verified' profiles is reliable⚠️
Please retweet 🙏
parisneuro.fr
doi.org
Jens-Bastian Eppler
Joao Barbosa
We find: A broad class of networks naturally does exactly this.
Mathematical intuition:
The same structure can be represented in different coordinate systems. A projection into a new basis will make activity look very different…
…while preserving the geometry of the population code.
3/5 🧪🧠
Jens-Bastian Eppler
The idea is, the path between them is still built from small changes:
🔴 red → 🟠 orange → 🟡 yellow → 🟢 green → 🔵 blue
Preserve local relationships all along the path, and the global manifold is preserved too.
So, colour perception will always be represented by a ring-like structure.
5/5 🧪🧠
Experiments often find:
- activity of individual neurons changes over days/weeks → population activity “drifts”
- but the geometry of the neural manifold stays remarkably stable
So, how can activity change while structure is preserved?
That’s the question we wanted to understand.
2/5 🧪🧠
This will be fun.
Including my talk on how neuronal activity manifolds are preserved in random networks on Tuesday.
Hope to see some of you there!
A second intuition:
Networks tend to preserve small differences. If two stimuli are similar, their outputs stay similar.
Example: 🔴 red → 🟠 orange
Thus, even if activity drifts, nearby points remain nearby.
But what about very different stimuli, like 🔴 red vs 🔵 blue?
4/5 🧪🧠
Very cool work from former colleagues: doi.org/10.48550/arX...
They render 3D objects into 2D images, systematically varying parameters like hue, lighting, camera angle, etc.
Having tightly controlled stimulus sets like this will help compare representations across AI and biological networks.
🧪🧠
Jens-Bastian Eppler
Two of the authors are here on Bluesky:
@pamelaosuna.bsky.social
@martinagvilas.bsky.social
Congrats (of course also to the ones not on here)!
🧪🧠
