Misc. ALEV and Neural Network Analysis

There are estimations that there are more than 600 different mathematical-statistical methods for modeling, dealing with risk and indetermination. [Proske, 2006]

Data-based modeling is characterized through consideration of expert knowledge, observation data and measurements for mapping nonlinear system behaviour. Opposite to analytical approaches data-based models "learn" complex system dynamics.
The bio-inspired paradigm of Neural Networks (NN) is based on Neuro-Biology, Neuro-Informatics, System Theory and . A low memory demand and short recall times are hereby the essential advantages for near realtime applications using high dimensional data. Contrary to classic expert systems a NN is an associative memory, so that large data bases are not needed to be linked and queried in recall phase. With the use of special approaches for data reduction it is possible to shrink down very large data sets before model calibration is started. By this memory consumption and calculation times are reduced significantly.
In case that available raw data is few the density of patterns can be increased by simulation runs. The real and virtual data is then the basis for model training.
The images below show ALEV-visualizations for some data from the UCI Machine Learning Repository.


General areas of application in socio-technical and
socio-economic systems ("soft systems"):

  • Clustering,
  • Classification - "Pattern Recognition",
  • Prediction, and
  • Decision Support.

Some areas of application in engineering ("hard systems"):

  • Noise Filter,
  • Replacement of missing values,
  • Technical Processes (e.g. roller mills, grinding, milling and welding),
  • Production Lines,
  • Material Flows and Supply-Chains,
  • Handling Technology, and
  • Robotics.

The TopLab guides und supports you with:

  • System descriptions,
  • Data collection and -coding,
  • Modeling,
  • Benchmarking,
  • Analysis and Reporting,
  • Documentation and Validation.