Sand Kit contains libraries designed for neural networks training and calculations, and a Linux kernel module for the SAND board neuro accelerator. The calculation library supports Multi Layer Perceptron (MLP), Radial Basic Function (RBF), and Kohonen networks in hardware acceleration mode and MLP and RBF networks in software emulation mode. It provides a C++ class library and simple C routines for neural network calculations. It has a flexible configuration which includes setting the maximum permitted memory usage, calculation and memory usage optimization options, input and output data sources, and debugging options. The configuration can be changed simply by editing the configuration file and/or with the built-in Gtk+ user interface. The learning library provides a flexible way to write new learning algorithms and methods using a powerful modules library. It contains several common learning routines, including an enhanced random search and resilient backpropogation algorithms. It is integrated with the calculation library to use software as well as hardware modes. All methods have separate configuration files. Support for interface plugins is provided, and all learning routines have built-in Gtk+ interface plugins.
