FeeLab/seqNMF

An algorithm for unsupervised discovery of sequential structure

SeqNMF is an algorithm which uses regularized convolutional non-negative matrix factorization to extract repeated sequential patterns from high-dimensional data. It has been validated using neural calcium imaging, spike data, and spectrograms, and allows the discovery of patterns directly from timeseries data without reference to external markers.
For more information see our preprint: https://www.biorxiv.org/content/early/2018/03/02/273128