20世纪50年代末,Frank Rosenblatt等人提出了一种称为感知机的神经元网络。引入了用于训练神经网络解决模式识别问题的学习规则。证明了只要求解问题的权值存在,那么其学习规则通常会收敛到正确的网络权值上。整个学习过程较为简单,而且是自动的。只要把反映网络行为的实例提交给网络,网络就能够根据实例从随机初始化的权值和偏置值开始自动的进行学习。 In the late 1950s Frank Rosenblatt et al. Proposed a neural network called a perceptual machine. Learning rules for training neural networks to solve pattern recognition problems are introduced. It proves that as long as the weight of solving the problem exists, its learning rule usually converges to the correct network weight. The whole learning process is simple, but also automatic. As long as an instance reflecting the behavior of the network is submitted to the network, the network can automatically learn from the randomly initialized weights and offset values ​​based on the instance.