Neural network training method and device based on many-core system, and electronic equipment

• Main Authors: WANG GUANRUI, QU HUANYU, ZHANG WEIHAO
• Format: Patent
• Language: Chinese; English
• Published: 11.10.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention provides a neural network training method and device based on a many-core system, electronic equipment and a computer readable medium. The method comprises the following steps: respectively loading at least one forward node and at least one reverse node corresponding to the forward node into one or more processing cores of the many-core system; in the forward propagation process, the forward nodes are controlled to execute task processing based on the corresponding first learnable data; in the back propagation process, the reverse node is controlled to execute error reverse transmission based on the corresponding second learnable data, and the forward propagation process and the back propagation process are executed in parallel; and when a preset updating condition is satisfied, updating the first learnable data and the second learnable data to obtain a trained neural network. According to the embodiment of the invention, the operation performance of the neural network on a many-core chip can be