#demo1 This tutorial introduces new users to NeuroSolutions by constructing and training a familiar neural network, the multilayer perceptron (MLP), to solve the exclusive-or problem.
#demo2 Multi-Layer Perceptron, Advanced
#demo2 This tutorial provides a more detailed explanation about how to construct neural networks. It rebuilds the network of the first tutorial, one component at a time, and again uses it to solve the exclusive-or problem.
#demo3 Linear Associator
#demo3 This tutorial introduces linear associative memories for performing hetero-association. Here we train the network to associate the images of three different faces with the images of their respective names.
#demo4 Linear Adaptive Filter
#demo4 This tutorial presents the linear adaptive filter, also known as the Adaline, and trains it to convert a square wave into a sine wave, thus extracting spectral components of the input.
#demo5 Modular Neural Network
#demo5 This tutorial introduces modular neural networks, which employ parallel MLPs that are summed at the final layer. We apply it to the problem of classifying the five stages of sleep from real data obtained through biological measurements of a human during one night of sleep.
#demo6 Jordan-Elman Neural Network
#demo6 This tutorial uses Elman's topology to double the frequency of a sine wave. The Jordan-Elman network is a multilayer perceptron with context units attached to the input layer. Context units provide memory through feedback to themselves with a time constant that determines the memory depth.
#demo7 Gamma Neural Network
#demo7 This example presents the focused Gamma neural network (FGNN) to solve the same problem as with the Jordan-Elman network; doubling the frequency of a sine wave. The FGNN is a multilayer perceptron with a gamma memory structure at its input. The gamma memory is a locally recurrent infinite impulse response (IIR) filter with an adaptable memory depth.
#demo8 Time Lagged Recurrent Network
#demo8 This example presents the Laguerre neural network for predicting the Mackey-Glass chaotic time series. Like the Gamma neural network, the Laguerre network also uses locally recurrent infinite impulse response (IIR) filters as adaptable memory structures.
#demo9 Unsupervised Learning
#demo9 This tutorial introduces two of the most popular unsupervised networks: adaptive principal component analysis (PCA) and competitive networks. These structures are commonly used in preprocessing layers of hybrid network topologies. PCA linearly projects the input space onto a smaller dimensional space, while preserving maximum energy. Competitive networks quantize the input space around regions where the data is clustered.
#demo10 Kohonen Feature Maps
#demo10 This tutorial introduces Kohonen's self-organizing map (SOM). These maps are commonly used as vector quantizers or as preprocessors in a hybrid network. SOMs perform vector quantization such that neighbors in the input space are also neighbors in the output space.
#demo11 Radial Basis Function Network
#demo11 This example trains a radial basis function (RBF) network to monitor machine tool breakage. RBF's sum the activations from a layer of Gaussian axons, and thus they differ from MLP's in that their overall input-output map is constructed from local contributions. They train faster than MLP's, and require fewer exemplars.
#demo12 Principal Component Analysis Hybrid Network
#demo12 This example uses a hybrid PCA/MLP network for digit recognition. The idea is to use the unsupervised layer (PCA) at the input as a feature extractor, and the supervised layer (MLP) as a classifier.
#demo13 Support Vector Machine, Kernel Adatron
#demo13 This example uses a support vector machine (SVM) to classify two sets of data which share a complex boundary. The SVM is implemented using the kernel Adatron algorithm. The kernel Adatron is adept at identifying and utilizing boundary inputs to optimally classify sets of data with complex boundaries.
#demo14 Fuzzy Network - CANFIS
#demo14 This tutorial demonstrates the capability of a neural fuzzy inference system to approximate a non-linear sinc function. The CANFIS model integrates fuzzy inputs with a modular neural network to quickly solve poorly defined problems. Fuzzy inference systems are also valuable as they combine the explanatory nature of rules (membership functions) with the power of neural networks.
#demo15 Genetic Optimization
#demo15 This tutorial demonstrates the use of genetic algorithms to optimize neural network parameters. Through genetic optimization, virtually any network component parameter(s) can be probed to discover an overall optimal network configuration (as determined by a network fitness function).
#demo16 The Power of Macros
#demo16 This tutorial introduces NeuroSolutions' macro language. All of these demos were written with macros.
#demo17 OLE Automation
#demo17 This tutorial demonstrates communicating with NeuroSolutions from other programs. NeuroSolutions is an OLE server, and thus it can receive OLE messages and data from an OLE-compatible application and return data in response. The NeuralWizard, for example, sends OLE commands to NeuroSolutions in order to build the user-designed network.
#demo18 Code Generation
#demo18 This tutorial introduces code generation, which allows the user to generate ANSI-compatible C/C++ code for running simulations on other platforms.
#demo19 Dynamic Link Library Demos
#demo19 This link takes you to another menu page with five new demos introducing dynamic link libraries. DLLs allow the user to customize NeuroSolutions' components through simple C code modifications.
