Big Scientific Data and Machine Learning in Science and Engineering 

Scientific observations, experiments, and simulations provide valuable data to train machine learning algorithms for constructing surrogate models, also known as metamodels, that assist in characterizing complex systems. The adoption of machine learning surrogate models holds the potential to substantially accelerate design space exploration and optimization when a closed analytical form that relates input parameters to performance criteria is unattainable. Examples of such systems abound in a wide range of scientific problems involving multiple scales, such as climate and power grid modeling, biological simulations, structural engineering, and materials science. However, despite the enormous potential for exploiting data-driven surrogates, collecting and analyzing massive amounts of data poses significant challenges. Hence, there is a growing interest in focusing on the synergistic integration of traditional methods of scientific discovery and recent machine learning techniques, e.g., deep learning. Three fundamental research directions include: (1) assuring training data quality and adequacy under stringent resource constraints; (2) developing efficient, reliable, and physics-informed data-driven models to fully capture the behavior of complex systems; and (3) leveraging surrogate models for performing domain-specific tasks, such as sensitivity analysis and uncertainty quantification.

Topics of Interest
This special session invites submissions of original works that address the above and other unique challenges of developing machine learning surrogate models in a broad range of science and engineering applications. Topics covered by this special issue include but are not limited to: 

• General methods for data acquisition, exploration, and analysis
• Subset selection in machine learning 
• Active learning and optimal experimental design 
• Feature extraction, selection, and dimensionality reduction 
• Predictive modeling in the limited labeled data and/or weakly supervised case 
• Machine learning methods for multimodal and heterogeneous data sources 
• Online methods for streaming data
• Physics-informed machine learning 
• Developing surrogate models for complex systems and model evaluation 
• Exploration and optimization of design spaces with machine learning surrogate models 
• Multi-fidelity uncertainty quantification 

All manuscripts submitted must be original, not under consideration elsewhere, and not previously published. The peer-review process is designed to avoid bias and conflict of interest on the part of reviewers and is composed of experts in the relevant field of research. A key criterion in publication decisions will be the manuscript’s fit for the special issue. Papers will be published online as soon as accepted. 

Deadline for Submissions: April 1, 2022 

Submission Guidelines 
Prospective authors are requested to submit new, unpublished manuscripts for inclusion in the upcoming event described in this call for papers. Paper submissions for this special issue should follow the submission format and guidelines. Paper submissions for this special issue should follow the submission format and guidelines.