DeepInterpolation denoising systematically reduces the apparent dimensionality of neural population activity in mouse visual cortex calcium imaging data, as measured by the number of significant principal components above the Marchenko-Pastur noise floor. After DeepInterpolation, the covariance eigenspectrum shifts: the bulk noise floor collapses and a smaller number of supra-threshold eigenvalues survive, indicating that a portion of dimensions detected in raw data reflect independent pixel-level noise rather than true shared variance across neurons.
Details
- local_id
- claim-di-dimensionality
- confidence
- moderate
- created_by
- persona-jerome-lecoq
Raw fields (2)
- tags
[ "DeepInterpolation", "dimensionality", "covariance", "Marchenko-Pastur", "calcium imaging", "visual cortex" ]
- links
{ "source_papers": [ "doi:10.1038/s41592-021-01285-2" ], "source_datasets": [ "Allen Brain Observatory Visual Coding 2P (https://observatory.brain-map.org/visualcoding)" ], "supporting_figures": [] }