SIHH exhibitors member, the top watch brands in the industry with both Patek Philippe, Breguet under a high Vacheron Constantin, Audemars Piguet and Lange, and Van Cleef & Arpels and Piaget are replica watches the watchmaking and jewelry Aesthetics combined leader. On the rolex replica technical strength and the movement of manufacturing skills, few brand par with the replica watches Jaeger-LeCoultre, Parmigiani control technology to meet the people and minority tastes worldwide; Panerai and IWC as omega replica Richemont contribution classic flagship models, whether or Portugal Lumior series, table fans are replica rolex in the advanced stage of difficulty neglected works.

Neural Network Simulator

The Neural Network Simulator is a graphical simulator for neural networks running under Microsoft Windows, that allows to create networks of different types and unlimited size and later integrate them into custom applications. Among a wide range of available types the user can choose the most appropriate one to create his network. Networks of different types can be connected to share input and output signals. Among the supported net types are Single Layer Perceptrons (SLP), Multi Layer Perceptrones (MLP), radial basis function networks (RBF), recurrent networks (Elman and Jordan), bidirectional associative memory (BAM) and different varieties of self organizing maps (SOM). Furthermore the simulator has a special testing environment for creating biological networks without any mathematical constructs behind. There are no limitations of the size of the created networks. If one machine is not powerful enough to simulate the whole network then it is possible to connect several instances of the simulator that are running on different machines together to one cluster. It is also possible to export projects so that they could be loaded on other machines or to show them to other developers. A powerful API (Application programer interface) is accessible in order to include almost the whole functionality of the simulator into custom applications. Exported networks could be integrated into .NET application and used there in a productive environment. The simulator offers many different tools for analyzing the networks status. 2D and 3D diagrams can give hints on defects in the network.

Neural Network Editor

In the editor it is possible to add neurons (round objects) and connect them with each other. Neurons have different states according to their activity. The lines symbolize the connections between the neurons. Every line has a number assigned to it, which indicates the strength of the connection.

Biological network

Biological networks don't include any mathematical constructs. They transmit signals in the same way as neurons do in a human's brain. The users can excite neurons and watch how their signals are transmitted through the network.

Self organizing map (SOM)

This image shows the training process of a self origanizing map, which is used to recognize letters. On the left there is an input layer of 5x7 input neurons. The selection represents the letter 'A'.

Error landscape viewer

Whether a network produces the desired output depends on the weights of the connections of the neurons. For simple networks it is possible to draw maps, where the x and y axis symbolize possible weight values for two connections. The z axis displays the calculated error. Flat spots on the map indicate weight combinations with a low network error.

Function Manager

Every neuron translates it's input signals into an activation and it's activation into an output. How these translations are done depends on the used functions. The simulator contains a default set for the most common functions, but the user also has the option to define his own functions that are later assigned to the neurons.

Lesson Manager

Most networks are trained based on a predefined set of input patterns. These patterns could be created in the lesson manager or imported from other sources. It is also possible to export lessons. Every network can have one lesson assigned to it. This lesson is used as a source for the input signals of the network's input neurons.

Network settings

Every network has a range of parameters that could be configured to adjust the net to a given sitation.

Neuron settings

In the neuron settings the user can control how the internals of a neuron work. Translation of input signals into output signals is done by assigning functions.

New network dialog

In the new network dialog the user can choose among several different types of networks. After one type is selected a list of supported neuron types for that network appears and the user can specify the inital amount of each neuron type. Later neurons could still be added or removed from the network.

Net error viewer

The image shows the training process of a network. After every training step the network error is calculated. The error is displayed in a diagram that updates automatically during the training process.

Two dimensional input viewer

For self organizing maps with only two input neurons a special dialog could be used to watch the learning process of the net. The dialog shows the training patterns and the SOM neurons on one map. The closer the neurons get to the patterns, the better the network representation for these patterns is.

Demonstration Video