Many recent technological breakthroughs are only made possible by probabilistic learning methods. These cover a large span of areas of expertise, e.g. robotics, image recognition or web search algorithms. These techniques combine ideas from machine learning with probability theory.

Our patented solution Cepel is such a probabilistic learning technique. Cepel is an acronym for Constructive Probabilistic Learning. It distinguishes itself from other methods by harnessing in-depth specialized knowledge, thus substantially improving the quality of the results. This new approach to the construction of solutions provides very precise results in many different areas of expertise.

The kernel based estimator of Cepel uses a Set Adaptive Kernel. This way small but important details of the resulting distribution are maintained while the overall curve stays smooth.

The parameter selection employs the Q Function, which itself applies an extended form of cross validation to evaluate the quality of the estimation. A normalization neutralizes linear scalings. The extensive Model Notation and several methods for smart performance improvement complete the technique.