The convolutional neural network (CNN) architecture with convolutional layers and downsampling layers was introduced by Kunihiko Fukushima in 1980.[30] He called it the neocognitron. In 1969, he also introduced the ReLU (rectified linear unit) activation function.[31][7] The rectifier has become the most popular activation function for CNNs and deep neural networks in general.[32] CNNs have […]
