Cortical Modeling Software in Matlab¹

• Version 2.0.1 contains all of the data and code required to replicate the experiments described in Brown University Department of Computer Science Technical Report CS-05-13. (TAR) (README)
• Version 2.1.0 includes a parallel implementation of the algorithms described in Version 2.0.1 using MPI. This version relies on a modification of Jeremy Kepner's MatlabMPI which is included and therefore does not require any additional installation to run. (TAR) (README)
• Version 3.0.3 is the beta release for the first fully functioning version implementing spatio-temporal features. It includes a modified version of Kevin Murphy's Bayes Net Toolbox and a demonstration based on the pictionary dataset developed by Dileep George. This version was used as the basis for a C++ implementation² relying on Intel's open-source Probabilistic Networks Library. (TAR) (README)

Supplementary Libraries and Datasets

• This supplement explores the role of inertial priors in learning invariant features. (TAR)
• This supplement investigates the use of mixtures of Gaussians and tree-augmented naive Bayes networks³ for modeling simple cells in the striate cortex. (TAR)
• Since the level-one structures in our cortical models are learned in an unsupervised manner and are largely independent of the large-scale properties of the training data it is possible to reuse them in multiple experiments. This tarball includes cached versions of networks used on the pictionary dataset and that can be used to expedite training using the restore function. (TAR)