Connectionists: 2 year Post-Doc - Charles University in Prague - Bioinspired Recurrent Neural Network Architectures

[Jan Antolik]

Postdoctoral researcher - bioinspired RNN architectures – Charles
University in Prague


The CSNG Lab <https://csng.mff.cuni.cz/> at the Faculty of Mathematics and
Physics at the Charles University  is seeking a highly motivated *Postdoctoral
Researcher* to join our team to work on *digital twin model of visual
system. *Funded by the JUNIOR Post-Doc Fund <https://cuni.cz/UKEN-178.html>,
this position offers an exciting opportunity to conduct cutting-edge research
at the intersection of systems neuroscience, computational modeling and AI.


*Project description:*

Modern deep-learning techniques have transformed visual neuroscience by
substantially improving the ability of models to predict cortical neuron
responses to unseen visual stimuli. However, current deep-learning
methods have two major shortcomings. First, they focus on predicting only
the average neural response, failing to capture the fine temporal dynamics
generated within recurrent neural populations. Second, these models rely on
standard "off-the-shelf" architectures optimized for efficient training
rather than reflecting the biological substrate under study. As a result,
they function as black boxes, making it difficult to interpret the learned
representations in terms of how visual processing is organized and
implemented in biological neural circuits. These limitations hinder the
ability of
deep-learning models to provide meaningful insights into the principles
governing vision and translate them to  clinical applications such as
brain-machine-interface systems.

To address these challenges, in this project *we will develop novel
modular, multi-layer recurrent neural **network (RNN) architectures that
directly mirror the architecture of the primary visual cortex.* Our
models will establish a one-to-one mapping between individual neurons at
different stages of the visual pathway and their artificial counterparts.
They will explicitly incorporate functionally specific lateral recurrent
interactions, excitatory and inhibitory neuronal classes, complex
single-neuron transfer functions with adaptive mechanisms, synaptic
depression, and others. We will first train our new RNNs on synthetic data
generated by a state-of-the-art biologically realistic recurrent spiking
model of the primary visual cortex developed in our group. After we
establish the proof-of-concept on the synthetic data, we will translate our
models to publicly available mouse and macaque data, as well as additional
data from our experimental collaborators.

*What do we offer:*

We are the Computational Systems Neuroscience Group (CSNG)
<https://csng.mff.cuni.cz/> based at the Faculty of Mathematics and Physics
of Charles University, Prague. The main goal of our group is to identify
computations implemented in the neural system that underlies sensory
perception, as well as applying this knowledge to the design of stimulation
protocols for visual prosthetic systems. To that end, we build models of
visual systems at various levels of abstraction using a variety of
computational techniques including, but not limited to, machine learning
and large-scale biologically plausible spiking neural network
simulations. The position is *fully funded for 2 years, *and comes with a
salary equivalent to ~2400 EUR/month. We offer a dynamic international
working environment and collaborations with world-leading experimental labs
(Stanford, University of Pennsylvania, Institute de la Vision Paris etc.).


*Candidate profile:*
Strong background in modern machine learning techniques. Prior experience
with training recurrent neural networks, and neuroscience or related
disciplines is an advantage, but not strict requirement.

Interested candidates should *contact Dr. Ján Antolík* (
jan.antolik at mff.cuni.cz) with their CV. *Deadline 5th August 2025.*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20250614/d0472e6b/attachment.html>