Unsupervised discovery of temporal sequences in high-dimensional datasets

Unsupervised discovery of temporal sequences in high-dimensional datasets

Date Posted: 

May 9, 2018

Date Recorded: 

April 19, 2018

CBMM Speaker(s): 


Andrew Bahle
  • Computational Tutorials


Emily Mackevicius, Andrew Bahle - MIT

The ability to identify interpretable, low-dimensional features that capture the dynamics of large-scale neural recordings is a major challenge in neuroscience. Dynamics that include repeated temporal patterns (which we call sequences), are not succinctly captured by traditional dimensionality reduction techniques such as principal components analysis (PCA) and non-negative matrix factorization (NMF).  The presence of neural sequences is commonly demonstrated using visual display of trial-averaged firing rates. However, the field suffers from a lack of task-independent, unsupervised tools for consistently identifying sequences directly from neural data, and cross-validating these sequences on held-out data. 

In this tutorial, we will describe a tool for unsupervised discovery of temporal sequences in high-dimensional datasets.  This tool, which we call seqNMF, extends a convolutional NMF technique.  It provides a framework for extracting sequences from a dataset, and is easily cross-validated to assess the significance of each extracted factor.  After describing how seqNMF works, we will demonstrate the application of seqNMF to several neural and behavioral datasets, and provide demo code.  You're encouraged to bring data and questions.