Brain time-space network decomposition method and system based on microneural structure search

Abstract

The invention discloses a brain time-space network decomposition method and system based on micro-neural structure search. The method comprises the steps: determining an input feature matrix, an output layer matrix and a target output matrix according to a brain original data matrix; according to the input feature matrix, the output layer matrix and the target output matrix, carrying out optimal search on a cell structure to obtain an optimal cell structure; introducing an early stop mechanism, and determining a final cell structure according to the optimal cell structure; and decomposing thebrain time-space network according to the final cell structure to obtain time dynamic characteristics and space network characteristics. According to the method, internal nodes in an RNN layer cell structure are introduced into a differentiable continuous search space for search updating, an early stop mechanism is introduced to construct a final cell structure, and a brain time-space network is optimally decomposed according to the final cell structure; therefore, a basis is provided for follow-up precision medical treatment, intelligent risk prediction and personalized education of brain diseases.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence brain network decomposition, in particular to a brain space-time network decomposition method and system based on differentiable neural structure search. Background technique [0002] Brain functional network is an important basis for the study of brain cognition and thinking process, and can provide important support for smart education, precision medicine and other fields. Recurrent neural network (RNN) has been widely used in the dynamic analysis of brain networks due to its natural advantages in describing the time dimension. However, limited by the structure of the current algorithm model itself, the current analysis and research on brain networks is very dependent on the prior knowledge of specific independent tasks, and it is impossible to generalize and analyze all scenarios in a specific field, such as: different cognitive tasks , different individuals, etc. Therefore, i...

AI Technical Summary

Problems solved by technology

However, the above differentiable neural network structure search framework has two defects: 1. It only considers the data analysis in the time dimension, ignoring the mining of space dimension and space-time interaction information, so it cannot achieve adaptive time-targeted - The RNN cell structure decomposed by the empty network is optimally searched; 2. The above search framework has a collapse issue (collapse issue), which will cause the cell structure to be too shallow, resulting in poor expressiveness

Images