Neural Phasors: A Neural Network And Image Segmentation Toolbox

This repository contains code that was written as part of my PhD thesis at the Institute of Cognitive Science. The following is a short list of some of the functionality that is implemented in this toolbox:

• A Sparse Autoencoder for unsupervised learning of features from natural images.
• Stacking several autoencoder layers to create a deep network.
• Code for ZCA-whitening of color images.
• Neural network of Kuramoto oscillators that are used for image segmentation.
• A complex-valued autoencoder implemented in the machine-learning framework Torch.
• Code to extract functional brain connectivity from EEG data using different types of coherence and causality measures.
• Probabilistic fiber tracking using diffusion tensor imaging (DTI) data.
• A GUI for paramter exploration on the Sun Grid Engine.
• Dynamic neural network simulations of Jansen-Rit neural mass models in the cortical connectome.

Instructions to run the code:

Edit the paths in "matlab/include/dataPaths.m" to match your system setup.

In Matlab run the following script to add all relevant paths to the environment: ''' addScriptPaths(); '''

There are several scripts in the parameters subfolder that can be used to start computations.

List of publications that were produced using code in this repository:

• Finger, H., Gast, R., Gerloff, C., Engel, A. K., & König, P. (in review). Probing Neural Networks for Dynamic Switches of Communication Pathways. PLoS Cb. Link

• Finger, H. (2017). Information Process- ing in Neural Networks: Learning of Structural Connectivity and Dynamics of Functional Activation. Dissertation. Link

• Finger H, *Bönstrup M, Cheng B, Messé A, Hilgetag C, Thomalla G, et al. (2016). Modeling of Large-Scale Func- tional Brain Networks Based on Struc- tural Connectivity from DTI: Compari- son with EEG Derived Phase Coupling Networks and Evaluation of Alternative Methods along the Modeling Path. PLoS Comput Biol 12(8): e1005025. doi:10.1371/journal.pcbi.1005025. Link

• Finger, H., & König, P. (2014). Phase synchrony facilitates binding and segmentation of natural images in a coupled neural oscillator network. Frontiers in computational neurosci- ence, 7, 195. Link