Algebraic analysis of neural net learning

Summary form only given, as follows. An approach is presented to the algebraic analysis of learning paradigms in neural nets. The technique is to map the learning paradigm into a learning automaton with certain convergence characteristics. Such automata have been studied by the authors, and their algebraic structure was analyzed. From this structure a lower bound can be assigned to the number of steps in a learning sequence. Using the mapping, a similar lower bound can be deduced for the learning paradigm.<>