Computation of multiweight
K. Maitree, S. Kondo · Electronics Letters · 1980
In learning machines the key problem is to determine weight vectors satisfying linear inequalities. In two-category classification only one weight is needed, but in multicategory classification the weight vectors which have to be found are as many as the number of pattern classes. The letter shows that, for linear machines, it is necessary to find only one weight vector, which the authors call the multiweight, by applying the steepest descent method in a learning algorithm.