A Polynomial Time Algorithm for the Construction and Training of a Class of Multi-Layer Perceptrons

该奖项资助了一类用于分类问题的多层类感知器神经网络的研究。该算法采用线性规划方法，在有限的高阶感知中增量生成隐藏层，在多项式时间内可训练。该研究将研究该算法的扩展，以解决大型训练集的分类问题，这些问题通常需要很长时间才能收敛。该算法的行为将在计算机视觉和机器人的代表性问题上进行评估。

This award funds research on a class of multilayer perceptron- like neural networks for classification problems. The algorithm uses linear programming methods to incrementally generate hidden layers in a restricted higher-order perception, which is trainable in polynomial time. The research will investigate extensions of the algorithm to address classification problems with large training sets, which normally take a long time to converge. Behavior of the algorithm will be evaluated on representative problems from computer vision and robotics.