认知中的联结主义模型

连接主义模型是作为一种受大脑启发而提出的，作为符号模型的替代方案。一个典型的连接主义网络由多层节点组成：输入层、一个或多个隐藏层以及输出层。节点之间的每个连接都有一个数值权重，该权重可以是兴奋性的，也可以是抑制性的。当一个模式呈现给输入层时，激活信号会沿着网络传播，并根据每个节点的权重和激活函数进行调整，最终在输出层产生相应的模式。

The key novelty is the learning process. In supervised learning, the network’s output is compared to a target output, and the difference (error) is used to modify the connection weights throughout the network. The backpropagation algorithm is a common method for efficiently calculating these weight adjustments. This process allows the network to gradually ‘learn’ complex mappings from inputs to outputs without being programmed with explicit rules. This approach excels at tasks involving noisy data and pattern recognition, such as object recognition or learning the past tense of English verbs, which are challenging for purely symbolic systems.