1. mackelab
  2. public
  3. pop_spike_dyn




More information at www.mackelab.org

This package is written by:

  • Lars Buesing (primary contact), lars@stat.columbia.edu

  • Jakob Macke

  • Yuanjun Gao

This repository contains different methods for linear dynamical system models with Poisson observations. It has been developed and implemented with the goal of modelling spike-train recordings from neural populations, but at least some of the methods will be applicable more generally.

In particular, the repository includes methods for

  • Laplace approximation for state-inference
  • Variational inference for the state
  • Expectation maximisation for parameter learning, using Laplace or Variation inference
  • Nuclear-norm minimisation (as described in Pfau et al.)
  • Exponential family PCA
  • Nonlinear Subspace Identification (SSID) (core code available at http://bitbucket.org/larsbuesing/ssidforplds)


To get started, run the example script: ./example/PLDSExample.m or one of the other scripts in ./example

If you notice a bug, want to request a feature, or have a question or feedback, please make use of the issue-tracking capabilities of the repository. We love to hear from people using our code-- please send an email to info@mackelab.org.

The code is published under the GNU General Public License. The code is provided "as is" and has no warranty whatsoever.


The code is based on

Jakob H Macke, Lars Buesing, John P Cunningham, M Yu Byron, Krishna V Shenoy, and Maneesh Sahani. Empirical models of spiking in neural populations. In NIPS, pages 1350–1358, 2011.

download paper

Lars Buesing, Jakob H Macke, and Maneesh Sahani. Spectral learning of linear dynamics from generalised-linear observations with application to neural population data. In NIPS, pages 1691–1699, 2012.

download paper

David Pfau, Eftychios A Pnevmatikakis, and Liam Paninski. Robust learning of low-dimensional dynamics from large neural ensembles. In NIPS, pages 2391–2399, 2013.

Yuanjun Gao, Lars Buesing, Krishna V Shenoy, John P Cunningham. High-dimensional neural spike train analysis with generalized count linear dynamical systems. In NIPS 2015

download paper

and some of the methods are also described in

JH Macke, L Buesing, M Sahani: Estimating state and model parameters in state-space models of spike trains. Book-chapter, in preparation.

The code-package makes use of the optimisation-package minFunc, written by Mark Schmidt, http://www.di.ens.fr/~mschmidt/Software/minFunc.html.