contributions (1)
a model architecture and model estimation algorithm for function approximation
the elements of the algorithm aren’t brand new, but it combines many nice features in a single framework and level of description.
CWM
- is general enough to handle many classes of problems.
- Is specific enough to be quickly applicable.
- is transparent.
- reuses approved past practice in the form of local models.
- converges quickly.
- allows for data statistics and uncertainties to be derived from the model.
- is easily extended into more complex architectures such as cluster-weighted sampling, cluster-weighted hidden-Markov models …