Category Intelligent Software>Neural Network Systems/Tools

Abstract ECANSE (Environment for Computer Aided Neural Software Engineering) provides an extensive development environment for the design, simulation and testing of Neural Networks (NN) and their applications. ECANSE provides a flexible environment and therefore it is suitable for a variety of purposes including:

1) Introduction to Neural Networks;

2) Designing Neural Networks; and

3) Designing Applications of Neural Networks.

ECANSE supports all design phases from the definition of a network up to the production of an optimized software solution. Product can be used as a design tool, simulation tool, process control tool, data analysis tool, forecast tool, rapid prototyping tool, visualization tool, or as a platform to compare different approaches.

ECANSE principles/components include:

Object Orientated -- Object Oriented Approach and Object Oriented techniques allow complex software systems to be designed and managed effectively (Software Engineering Aspect). The system is described in terms of many interacting objects organized in a hierarchy. Easy extension by the end-user and reusability are the main advantages of this approach.

Standardization -- ECANSE is based on industrial standards for software development tools such as Windows NT, C++, and SQL. A Standard Look-and-Feel is also supported.

Open System -- Open System Design and Easy Extension Object Oriented design and well structured objects make ECANSE an expandable design tool. New functions (e.g. new learning paradigms, statistical evaluations, data interface formats) can be integrated as feature packages. Thus, ECANSE can be configured by any user to fit their personal requirements by using the Programmer's Version of ECANSE.

Universality -- ECANSE offers a general development system for neuro- computing, Not just for specific applications. Different levels of abstraction (hierarchies) facilitate the handling of complex systems.

Graphical User Interface (GUI) -- A user-friendly graphical interface provides a practical and easy platform for designing and testing network models and their applications.

Component Object Model (COM) Connectivity -- ECANSE instances can be spread over several personal computers (PCs) that can control each other. Parameterization and attribute configuration can be done from every node with help from the ECANSE Object Designer.

COM Interfaces -- Many ECANSE Classes export their functionality via COM Interfaces.

ECANSE Objects -- ECANSE includes a wide range of objects or modules for:

1) Reading/writing data: e.g. American Standard Code for Information Interchange (ASCII), Sink, ODBCWrap, ExcelIn, ExcelOut;

2) Generating common deterministic and stochastic signals: e.g. Signal, Random, Chaos;

3) Preprocessing data: e.g. Filter;

4) Common mathematical operations and functions;

5) Statistics;

6) Control flow: e.g. Index, Tapped Delay Line (TDL), Time Delay, Multiplexer (MUX), Demultiplexer;

7) Visualizing data and monitoring performance: e.g. Graph, Graph3D, Color Map;

8) Creating sub-configurations: Configuration;

9) Neural Networks, Fuzzy Logic, Genetic Algorithms (GA);

10) Analysis: Mutual Information, Confidence Measure, Principle Components, and Descriptive Statistics.

ECANSE Scripts -- Each object can include a script which can be used to modify and customize the functioning of the existing objects in ECANSE.

Graphical User Interface --

Iconic Representation -- All of the ECANSE objects such as neural networks, file input/output, visualizers are represented graphically by icons. The user operates in a mouse driven windows environment. An object is placed in the configuration window by dragging & dropping the icon from the toolboxes. The user can also move the icon to any other position with the help of the drag & drop mechanism.

Parameterizing Objects -- By double-clicking on the icon, a dialog box opens where the user can change predefined parameters of the object. Furthermore, the user can assign any parameter/variable of the object to be input or output.

Connecting Objects -- The data flow is defined by the connectors. The objects transmit data to other objects by connecting the corresponding outputs to the inputs. Connecting objects together forms a configuration, which can also be connected to other objects and/or configurations forming hierarchical configurations.

Simulation Monitoring -- Parameter tuning is supported by specific objects that graphically represent and statistically evaluate the network performance. The understanding of network behavior when learning is simplified through the graphical visualization of the learning process in terms of the performance criteria, the learning rate, the weights, and any other parameters of interest.

Extending ECANSE -- ECANSE provides an ideal environment for the user to extend ECANSE with custom algorithms. A user can program their own customized objects into the ECANSE environment, in order to take advantage of the generic mechanisms such as parameterizing the object, connecting the object with other objects to form configurations, and simulating the configuration. The extension of ECANSE is enabled by the Programmer's Version of this product.

The following additional features/modules are included:

1) Multilayer Perceptrons and Back propagation Learning Hopfield Network, Associated Storage;

2) Radial Basis Function (RBF) Networks, Kohonen;

3) ScaLAPACK (Linear Algebra PACKage) Optimization;

4) Learning Vector Quantization (LVQ), Topological Feature Maps; and

5) Sensitivity Analysis.

System Requirements


Manufacturer Web Site ECANSE

Price Contact manufacturer.

G6G Abstract Number 20112

G6G Manufacturer Number 102360