Perceptron-based learning algorithms

摘要: A key task for connectionist research is the development and analysis of learning algorithms. An examination made several supervised algorithms single-cell network models. The heart these pocket algorithm, a modification perceptron that makes well-behaved with nonseparable training data, even if data are noisy contradictory. Features include speed fast enough to handle large sets data; scaling properties, i.e. methods scale up almost as well models when number inputs increased; analytic tractability, upper bounds on classification error derivable; online learning, some variants can learn continually, without referring previous winner-take-all groups or choice groups, be adapted select one out possible classifications. These suitable applications in machine pattern recognition, expert systems. >