Adam-based optimization method, system and terminal

Abstract

The invention discloses an Adam-based optimization method, system and terminal. The method comprise the following steps: inputting a training sample into a to-be-optimized Adam model with one or moremodel parameters to obtain a prediction result; comparing the prediction result with the real result, and respectively combining the prediction result with the size information of each model parameterunder the historical parameter gradient to obtain the size information of each model parameter under the updated gradient; obtaining a correction coefficient corresponding to the optimized Adam modelaccording to the training sample and the learning intensity value related to the optimized Adam model; obtaining a correction update amount of each model parameter so as to obtain an Adam optimization model used for obtaining an optimization prediction result. The method is used for solving the problem that the requirements of high accuracy and higher convergence rate of an algorithm cannot be met in the prior art for huge data volume. According to the method, model parameters are improved on the basis of an Adam model, and the convergence rate and the final accuracy are further remarkably improved while the excellent performance of the Adam is reserved.

Description

technical field [0001] The invention relates to the field of model parameter optimization, in particular to an Adam-based optimization method, system and terminal. Background technique [0002] The solution of model parameters is inseparable from the use of the optimizer in many cases. The quality of the optimization algorithm has a very important impact on the performance of the final model. In recent years, with the development of machine learning, especially deep learning, the demand for optimization algorithms for complex nonlinear and non-convex model parameters has been increasing. People's research on this aspect is also constantly deepening. As the model becomes more and more complex, higher requirements are put forward for the convergence speed and optimization effect of the optimization algorithm. In addition, the optimization algorithm can also help people better understand the learning process of the model, which is of great significance to the interpretabilit...

AI Technical Summary

Problems solved by technology

[0004] In view of the shortcomings of the prior art described above, the object of the present invention is to provide an optimization method, system and terminal based on Adam, which is used to solve the problem of high accuracy and faster algorithms that cannot be satisfied by the prior art for a huge amount of data. The problem of the requirement of the rate of convergence

Images