People don't always start from scratch when they think, and the results of past thinking will provide some support for future thinking. For example, in a conversation, a sentence means to be understood in context, not from scratch. Traditional neural networks cannot do this. For example, while convolutional neural networks can classify images, they may not be able to analyze the associations of each image in a video. The information of the previous image cannot be used for the analysis of the latter image, and recursive neural networks can solve this problem. The sequence data is processed by recursive neural network. Sequence data is data collected at multiple points in time, indicating the state or extent to which a thing or phenomenon changes over time. However, the sequence data has a characteristic, that is, the latter is related to the previous data. Traditional neural networks can only establish weight connections between layers, while recursive neural networks can also establish weight connections between neurons.
正在翻译中..