Konuşmacılar
Açıklama
This paper introduces a machine learning-based solution for efficiently solving sparse linear systems, a prevalent issue in scientific and engineering domains. The proposed approach leverages machine learning to predict the optimal number of partitions required for the block Cimmino method, a parallel hybrid solver to solve these systems. Traditional trial-and-error methods for determining the best number of partitions are surpassed by the proposed machine learning model, which is trained and validated on a diverse dataset from the Suite Sparse Matrix Collection. The model's performance is enhanced through feature selection and the use of Principal Component Analysis (PCA) for dimensionality reduction during training. The optimal model is selected based on the validation score and is evaluated using accuracy, AUC, and F1-Score metrics for both the training and testing phases. This method offers a faster and more efficient alternative to traditional approaches, with potential applications in various numerical methods in science and engineering. The study underscores the importance of appropriate feature selection for training the model and suggests future work to broaden the dataset and investigate other machine learning models to further improve the accuracy and efficiency of the proposed solution.