For years, physics and artificial intelligence have been seen as two separate scientific endeavors and have pursued different tasks almost independently. Physical models and solutions are interpretable but can have limited performances, are human labor intensive, and sometimes require several iterations to converge on an acceptable solution. In contrast, in the last years, artificial intelligence has made enormous progress in several fields such as prediction, classification, image processing, and regression problems. Unfortunately, artificial intelligence tools produce outputs that are usually not interpretable, lack physical basis, are problem specific, and therefore provide a limited contribution to the understanding of the phenomena under study and to the cumulative growth of knowledge.
Recently, there has been a relevant effort in combining artificial intelligence and physics to develop new performant and explainable algorithms. Explainable Artificial Intelligence (XAI) and Explainable Machine Learning (XML) allow the development of new solutions that are easy to interpret and allow scientists to identify and interpret the best model for solving a specific problem. The new Physics-Informed Neural Networks (PINN) have paved the way for new methods for numerical simulations, inverse problem solutions, PDE model discoveries, and much more. Genetic programming combined with symbolic regression allows fitting data with meaningful equations making recourse to a minimum of a priori assumptions. A series of data-driven methods have also been recently devised to go beyond pure statistical correlations and to identify the actual causal relations between the systems under study.
This Research Topic aims at collecting high-quality contributions to highlight the potential of these new methodologies in the natural and technical sciences, such as physics and engineering. The collection welcomes, but is not limited to, contributions to the following topics:
1. Innovative Physics-Informed Neural Network architectures and training methods applied to scientific problems.
2. Solution of complex PDE by using Artificial Intelligence and Physics-Informed Neural Networks.
3. AI-based inverse problem solutions.
4. AI-based model discovery for physics fidelity.
5. Physics-Informed and Explainable Artificial Intelligence tools applied to classification, regression, prediction, and clustering problems.
6. New genetic algorithms and their applications to solving scientific problems.
7. New methodologies and procedures to interpret machine learning and deep learning decisions and logic in science.
8. Machine learning identification of models for feedback control.
9. Machine intelligence for causal detection and modeling.
For years, physics and artificial intelligence have been seen as two separate scientific endeavors and have pursued different tasks almost independently. Physical models and solutions are interpretable but can have limited performances, are human labor intensive, and sometimes require several iterations to converge on an acceptable solution. In contrast, in the last years, artificial intelligence has made enormous progress in several fields such as prediction, classification, image processing, and regression problems. Unfortunately, artificial intelligence tools produce outputs that are usually not interpretable, lack physical basis, are problem specific, and therefore provide a limited contribution to the understanding of the phenomena under study and to the cumulative growth of knowledge.
Recently, there has been a relevant effort in combining artificial intelligence and physics to develop new performant and explainable algorithms. Explainable Artificial Intelligence (XAI) and Explainable Machine Learning (XML) allow the development of new solutions that are easy to interpret and allow scientists to identify and interpret the best model for solving a specific problem. The new Physics-Informed Neural Networks (PINN) have paved the way for new methods for numerical simulations, inverse problem solutions, PDE model discoveries, and much more. Genetic programming combined with symbolic regression allows fitting data with meaningful equations making recourse to a minimum of a priori assumptions. A series of data-driven methods have also been recently devised to go beyond pure statistical correlations and to identify the actual causal relations between the systems under study.
This Research Topic aims at collecting high-quality contributions to highlight the potential of these new methodologies in the natural and technical sciences, such as physics and engineering. The collection welcomes, but is not limited to, contributions to the following topics:
1. Innovative Physics-Informed Neural Network architectures and training methods applied to scientific problems.
2. Solution of complex PDE by using Artificial Intelligence and Physics-Informed Neural Networks.
3. AI-based inverse problem solutions.
4. AI-based model discovery for physics fidelity.
5. Physics-Informed and Explainable Artificial Intelligence tools applied to classification, regression, prediction, and clustering problems.
6. New genetic algorithms and their applications to solving scientific problems.
7. New methodologies and procedures to interpret machine learning and deep learning decisions and logic in science.
8. Machine learning identification of models for feedback control.
9. Machine intelligence for causal detection and modeling.
