Improvement of Computer Neural Network based on Feedforward Sequence Memory

The existing neural networks mainly involve working memory in the memory module, which can basically meet the needs of a neural network at work. However, for human intelligence, working memory is far from meeting the needs, especially in some advanced activities, such as thinking, language and emotion. Based on the concept driven processing theory, that is, the top-down processing mode of human perception, and the dualprocess theory of attention, that is, automatic processing and control processing based on the cognitive resource theory, this paper puts forward the computer neural network long-term memory storage model and the computer neural network dualprocess model, in order to construct the long-term memory of computer neural network, reduce the loss of computing time and computing resources, and improve the processing rate of information. It provides the possibility for the realization of neural network in automatic processing, subconscious judgment and other activities, as well as high-level activities such as self-control, strategy formulation and organization planning.