Thinks

TM THINKS^{^{^Ô}}

Neural Networks for Windows

Your Personal Neural Network Development Environment

Intuitive, interactive and highly responsive user interface

Built-in spreadsheets for convenient editing of data

Dynamic graphing and data displays continuously illustrate training progress

Industry standard, documented, field-proven algorithms, make THINKS the perfect tool for students and professionals

Ideal for forecasting, prediction, signal processing, classification, function approximation, decision making, sensor interpretation, and nonlinear regression

¨ THINKS provides maximum flexibility in six categories

Preprocessing Functions
Interconnection Architectures
Processing Element Summation Functions
Processing Element Transfer Functions
Learning Algorithms
Learning Error Criteria

¨ Preprocessing Functions

Thinks provides special input layer preprocessing to simplify working with all forms of data.

Min/Max and Mean/StdDev Input Preprocessing
Sum Inputs to 1 Normalization
Sum of Squares to 1 Normalization

¨ Processing Element Summation Functions

The function can be chosen independently on each layer of the network.

Dot Product Quadratic Sum
L1 Distance L2 Distance
Radial Basis Function Sigma-Pi

¨ Processing Element Transfer Functions

Thinks offers greater efficiency and training with 14 transfer functions.

Sigmoid
Bipolar Sigmoid
Arctan
Bipolar Arctan
Sin
Bipolar Sin
Threshold Linear
Bipolar Threshold Linear
Threshold
Bipolar Threshold
Linear
Gaussian
Cauchy
Winner Take All

¨ Interconnection Architectures

Processing elements can be connected in networks up to five layers.

Multilayer Normal Feed Forward
Multilayer Full Feed Forward
Total Recurrent
Cascade
Prior Recurrent
Cascade Recurrent

¨ Learning Algorithms

Thinks supports eight industry standard algorithms:

Back Propagation
Quick Propagation
Jacob's Enhanced Back Propagation
Recurrent Back Propagation
Kohonen Winner Take All
Kohonen Learning Vector Quantization
Simulated Annealing
Cascade Correlation Learning

¨ Learning Error Criteria

Control the network’s sensitivity to outliers in the training data with five error function options.

Mean Squared Error
Mean Absolute Error
Mean Fourth Power Error
Hyperbolic Square Error
Bipolar Hyperbolic Square Error

Thinks takes the confusion out of the choice of options and training by providing demonstration training files and examples. Each option is clearly described in on-line help files. The documentation includes a run-time library reference for each of the available functions.

Containing some of the most sophisticated algorithms and capabilities available today, THINKS provides a window into how neural networks learn and work. With THINKS, you can be training a network in minutes. Logical Designs also provides neural network software, for building custom applications, in several additional forms:

¨ TrainDos - PC/DOS executable program

Use this 32 bit extended DOS program for training of neural networks. The simple console interface and batch file processing make TrainDos ideal for many applications.

¨ Windows Dynamic Link Library (DLL)

Allows the user to build Custom Windows applications and run networks using C++, Visual Basic or Excel 5.0.

¨ Object Code Run Time Library

Allows the user to call all network subroutines from the user’s custom software.

To Order the Software…

Home

THINKS, TrainDos are trademarks of Logical Designs Consulting, Inc.
Windows, Visual Basic, Excel are trademarks of Microsoft Corporation.

Logical Designs Consulting, Inc. 2015 Olite Court, La Jolla, CA 92037 (858) 459-6236