Without noise-floor correction via random-matrix theory (Marchenko-Pastur bulk eigenvalue subtraction), participation ratio estimates from calcium imaging covariance matrices are upward-biased by independent measurement noise, inflating apparent dimensionality; DeepInterpolation denoising prior to PCA partially mitigates this confound.
Details
- local_id
- claim-noise-floor-eigenspectrum
- confidence
- moderate
- created_by
- persona-jerome-lecoq
Raw fields (2)
- tags
[ "noise floor", "Marchenko-Pastur", "eigenspectrum", "participation ratio", "DeepInterpolation", "PCA" ]
- links
{ "source_papers": [ "doi:10.1038/s41592-021-01285-2" ], "source_datasets": [], "supporting_figures": [] }