|
| |
cVision
Technologies
NGS’ technology is primarily based on an algorithm that supports
highly flexible network architectures and on the
introduction of innovative technologies into
neural networking. e.g. the development of the
completely connected perceptron (CCP). The effort for a successful and
efficient neural network design and training reduces dramatically. The implementation of
- Easy Handling of Numerical as well as Categorical Data
- Heuristic Network Initialization
- Intelligent Input Neurons
- Local Adaptive Learning Rules
- Heuristic Approach to Network Size
- Cluster Learning
- Superior Stopping Criteria
- Automatic Multi Fold Cross Validation
- Outlier Insensitive Networks
- Automatic Network QC
enables the design of almost every network architecture
for any task and ensures several advantages. The experts can concentrate on
their real problem and need not deal with network design, parameter tuning,
data normalization and all these in most cases extremely time consuming tasks.
Questions about optimal learning rates, number of hidden layers and hidden
neurons, network initialization and a lot of others are managed fully automatically
and let you focus on your problems rather than on the technology behind.
cVision
offers of course multi layer perceptron architectures with and without short-cut
connections, the number of hidden layers is arbitrary and the number of hidden
neurons is unlimited. In addition the completely connected perceptron is a more
general class of perceptrons embracing all possible multi layer perceptrons in it
with the advantage that a heuristic
approach to an optimal network size is really efficient and accurate.
|
