Abstract
Scientific discovery is rapidly evolving through the synergistic integration of traditional methods and modern machine learning techniques. Machine learning algorithms, trained on data from scientific observations, experiments, and simulations, are instrumental in constructing surrogate models. These models offer valuable insights into complex systems where a direct analytical relationship between input parameters and performance criteria is elusive.
Despite the immense potential of data-driven surrogate models, challenges persist in collecting and analyzing the vast datasets required for training. Therefore, current research efforts focus on three key areas. The first is ensuring the quality and adequacy of training data. The second is developing efficient and reliable models capable of accurately capturing the behavior of complex systems. Finally, the focus is on leveraging surrogate models for domain-specific tasks, such as sensitivity analysis and uncertainty quantification. By addressing these challenges, scientists are paving the way for a new era of scientific discovery that harnesses the power of both traditional methods and cutting-edge machine learning.
This Special Issue presents a collection of papers that aim to address these challenges.
Kriging, PCE and Low-Rank Approximations in Material Science and Big Data Analytics: This paper develops surrogate models to predict the nonlinear behavior of brittle materials such as concrete.
Hybrid generalized extreme learning machine through gradient-based optimizer model for self-cleansing nondeposition with clean bed mode of sediment transport: This paper utilizes a comprehensive experimental data and extreme learning machine for sediment transport modeling.
Vertical and Horizontal Water Penetration Velocity Modeling in Non-Homogenous Soil using Fast Multi-output Relevance Vector Regression: This paper studies a joint determination of horizontal and vertical movement of water through porous medium using multi-output relevance vector regression.
A New Filter Approach Based on Effective Ranges for Classification of Gene Expression Data: This paper presents new feature selection methods for analyzing gene expression data sets.
A Unified Training Process for Fake News Detection based on Fine-Tuned BERT Model: This paper develops a robust model to combine multiple data sets for enhanced fake news detection.
This Special Issue offers promising future research directions for creating accurate surrogate models and enhancing the quality of data sets used.