Dead zone training

In consequence of the randomness of the trained weight vector a classification by a committee machine (Chapter 2.6.2.) or a dead zone training is advisable C117, 123, 159, 2953. 

The dead zone training (Chapter 2.2.7) has also been used for feature selection. Those features are eliminated which do not affect... 

The program used for training, the flow diagram for which is given by Jurs et al. follows an error-correction procedure to arrive at acceptable values for the components of V. Initially aU the Fj components have -f 1 (or — 1) as their value. An initial dot product is then formed with the first mass spectrum in the training set. If the value of the dot product is outside the dead zone on the correct side, the response to the question, for example whether oxygen is present in the empirical formula, is deemed to be correct, and the next spectrum is processed. Whenever a spectrum is wrongly classified, a correction is made to F a ft multiple of the vector S is added to the old vector F in order to form a new vector F . 

FIGURE 20. Decision plane with thickness 2t. The training process tries to find a decision plane with no patterns within a dead zone of 4% from the plane. 

If the classifier is applied to unknown patterns, the dead zone from the training can be used as a rejection zone Patterns which give a scalar product between -t and +t are not classified- Chemical applications showed that the predictive ability for the remaining patterns outside the rejection zone is increased by this method C3203- The distance to the decision plane may be used as a measure of confidence for classifiers with a continuous response (Chapter 2.6.1.). 

---