Neuroinformatics Group
Universität Bielefeld › Technische Fakultät › NI
Search
Contributed Packages
The power of Neo/NST arises from its modularity. There are a plenty of additional packages available to extend the basic functionality. In the following only a few of them are listed. The full list is available on projects.cit-ec
Data Structures offers basic data structures (List, Graph, RingBuffer, Directory) as template classes for the prog-unit.
Competitive Layer Model allows for dynamical grouping of features exploiting a recurrent neural network dynamics.
ODBC and MySQL database support Allows to store and retrieve data from database servers. Two Neo units provide data access over ODBC function calls. Additionally a loadable library provides access to MySQL servers with Neo's prog_unit.
2D FFT image processing routines Fast Fourier transforms for image data with several preprocessing options. Including Gabor Jet feature vector computation and image restauration methods for focus and motion blur.
Hierarchical clustering package allowing for clustering based on single-linkage, complete-linkage, (weighted) group average, centroid, median or ward distance metrices.
A generalized SOM package which allows to "plug-in" arbitrarily customized embedding geometries for the SOM. Currently, spherical and hyperbolic geometries are readily available.
Vortex Simulation Toolkit provides a wrapper for the physics simulation Vortex, including dynamic scene loading, robot structures and kinematic solvers, etc.
Visualization Toolkit (VTK) provides a wrapper of the VTK Visualization Toolkit, an OO software framework for computer graphics, visualization, and image processing.
SVM and other kernel methods providing several algorithms utilizing kernel functions. Currently, the package contains a Support Vector Machine (SVM) unit for binary pattern classification and a unit for estimation of the support of some data.
Neural Network package provides methods to create, train, use and inspect neural network objects of arbitrary topology. Includes: backpropagation (MLP, Elman-, and Jordan nets) and backpropagation-through-time for general recurrent nets, Hebbian learning, delta rule.
Reinforcement Learning package implements class for Q-learning in discrete and continuous state and action spaces (continuous action space is currently limited to 1D only).
Interactable graphics objects package implements some frequently used graphics objects (such as rectangles, arrows, markers and marker_sets) with built-in mouse interaction.
- Login to post comments