Model performance reasoning method and device based on dynamic time sequence diagram

Abstract

The invention discloses a model performance reasoning method and device based on a dynamic time sequence diagram, and the method comprises the steps: mapping a training process of a neural network model into a time evolution diagram, extracting node features in the time evolution diagram, and further extracting multi-dimensional statistical features in the node features in order to predict the performance of a heterogeneous model. And aggregating the multi-dimensional statistical features into time sequence features to predict the performance of the model. The neural network model composition mode provided by the method is simpler than an expansion composition mode in an existing convolutional layer, the calculation efficiency is higher, and meanwhile, the performance of a prediction task is not sacrificed. According to the method, the neural network model performance prediction problem is solved by capturing the neural network dynamics in the early stage of the model training stage, the uncertainty of the neural network training result can be reduced, and the model training efficiency is improved.

Description

technical field [0001] The invention relates to the fields of artificial intelligence and image classification, in particular to a model performance reasoning method based on a dynamic sequence diagram. Background technique [0002] Neural networks advance many fields, including computer vision, image processing, natural language processing, and bioinformatics. As task complexity increases, networks become deeper and larger, requiring more computing resources, training data, and time. However, even if a lot of training resources and training time are spent, the final performance of the neural network is uncertain. For example, using an inappropriate learning rate in the model training phase makes the final performance of the model poor. [0003] Much big data is presented in the form of large-scale graphs or networks. Many big data with non-graph structure are often converted into graph models for analysis. The graph data structure is a good representation of the associat...

AI Technical Summary

Problems solved by technology

However, the research on neural network structures as graphs is relatively limited, and the structural changes of the underlying graph during training are mostly ignored in the literature.

Images