Convergence Round Prediction for Deep Learning Training Based on Machine Learning Algorithm
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
At present, deep learning technology has been implemented on a large scale in computer vision, natural language processing and other fields, and the efficiency of model training and the optimal allocation of resources have become the core research topics focused on by industry and academia. Machine learning algorithm is the core support of deep learning performance prediction task. By mining the intrinsic correlation between hyperparameters, network structure, hardware resources and other features and convergence rounds, the quantitative prediction of training effect can be realized. In view of the shortcomings of traditional algorithms, such as insufficient feature extraction ability and weak timing-dependent capture, this paper proposes HLOA-CNN-BIGRU regression algorithm, which first carries out correlation analysis and violin graph analysis, and then relies on various machine learning algorithms to carry out comparative experiments. The experimental results show that the proposed algorithm is superior to the comparative algorithm in all evaluation indexes, and its MSE, RMSE, MAE and MAPE indexes are the lowest of all models, the highest R² value, the smallest prediction error, the best fitting effect and generalization ability; Among the traditional algorithms, integrated algorithms such as GBDT and AdaBoost have similar performance and strong stability, while decision tree and linear regression algorithms have weak prediction performance. This research provides a feasible solution for accurate prediction of deep learning convergence rounds, which can effectively help improve model training efficiency and resource utilization.
Fields of Study & Subjects
Machine Learning and Data Classification
Advanced Neural Network Applications
Stochastic Gradient Optimization Techniques
Publication details
- Journal / source
- Applied and Computational Engineering
- Volume
- 228
- Issue
- 1
- Pages
- 92–98
- Publication date
- 30 March 2026
- Language
- EN