Off-manifold coding in visual cortex revealed by sleep
In the primary visual cortex (V1), neural activity display low-dimensional dynamics and complex sensory representations. By using activity during non-REM sleep, we revealed a low-dimensional intrinsic manifold structure, where movements and visual scenes are encoded. Surprisingly, visual scenes also engage an off-manifold subspace, which comprises of a the sparse activity of neurons that are strongly coupled by low-dimensional manifold. These findings highlight a key link between V1’s low-dimensional organization and sparse coding, showing how the brain can efficiently use the neural activity space. bioRxiv
Identifying patterns differing between high-dimensional datasets with generalized contrastive PCA
Comparing high-dimensional datasets are common in biology, but the appropriate methods to uncover distinct low-dimensional patterns remains challenging. Traditional dimensionality reduction methods only process single datasets, and existing tools are limited by the need for hyperparameter tuning and asymmetric handling of data. We developed generalized contrastive PCA (gcPCA), a hyperparameter-free method that symmetrically compares datasets and resolves these issues. PLOS Computational Biology
