Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: implicit model, generative modelling, neuronal morphologies, computational neuroscience, primary visual cortex
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Understanding the diversity and complexity of the morphology of different types
of neurons is important for understanding neural circuits. We need quantitative,
unbiased methods to capture the structural and morphological features of neurons.
With the advent of large-scale structural datasets, this analysis becomes feasible
using data-drive approaches. Existing generative models are limited to modeling
dendritic and axonal skeleton graphs, without considering the actual 3D shape. In
this work, we propose MORPHOCC, a model that represents the diversity of neu-
rons in mouse primary visual cortex (V1) in a single neural network by encoding
each neuron’s morphology into a low-dimensional embedding. From this embed-
ding the 3d shape can be reconstructed. We train our model on 797 dendritic
shapes of V1 neurons. The learned embedding captures morphological features
well and enables cell type classification into known cell types. Interpolating be-
tween samples in embedding space generates new instances of neurons without
supervision. MORPHOCC has the potential to improve our understanding of neu-
rons in the brain by facilitating large-scale analysis and providing a model for
representing neuronal morphologies.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7678
Loading