MS004 - Machine Learning Advances in Earth Sciences and Engineering
Keywords: architectures with hard constraints, digital twins, earth sciences, foundation models for physical sciences, GenAI for science, material science, multimodal data fusion, open-source scientific datasets/benchmarks
The "Machine Learning Advances in Sciences and Engineering" minisymposium aims to explore the integration of artificial intelligence and computational methods to unravel the complexities of natural systems, with a focus on earth and material sciences and engineering. These fields are characterized by significant model and parametric uncertainties, sparse or noisy data, and unpredictable conditions that challenge traditional deterministic approaches.
These sessions will spotlight cutting-edge advancements, including foundation models for physical sciences, data-driven architectures with hard constraints, and generative AI techniques for scientific discovery. They will also propose ways for tackling challenges related to applying digital twins to complex natural systems, where emerging concepts like “probabilistic twins” or “quantum twins” may offer more appropriate frameworks.
Furthermore, the sessions will provide a platform for discussing the development and application of open-source scientific datasets and benchmarks, essential for model validation and reproducibility. Special attention will be given to multimodal data fusion algorithms that integrate diverse data representations, enhancing the accuracy and robustness of models.
Differentiable programming, which allows the incorporation of physical laws and constraints into machine learning models, will be a key topic. This approach ensures that the resulting models are not only data-driven but also physically plausible and interpretable.
The minisymposium seeks to foster collaboration and drive innovation in computational methods, ultimately contributing to a deeper understanding and more effective solutions for complex scientific challenges.
These sessions will spotlight cutting-edge advancements, including foundation models for physical sciences, data-driven architectures with hard constraints, and generative AI techniques for scientific discovery. They will also propose ways for tackling challenges related to applying digital twins to complex natural systems, where emerging concepts like “probabilistic twins” or “quantum twins” may offer more appropriate frameworks.
Furthermore, the sessions will provide a platform for discussing the development and application of open-source scientific datasets and benchmarks, essential for model validation and reproducibility. Special attention will be given to multimodal data fusion algorithms that integrate diverse data representations, enhancing the accuracy and robustness of models.
Differentiable programming, which allows the incorporation of physical laws and constraints into machine learning models, will be a key topic. This approach ensures that the resulting models are not only data-driven but also physically plausible and interpretable.
The minisymposium seeks to foster collaboration and drive innovation in computational methods, ultimately contributing to a deeper understanding and more effective solutions for complex scientific challenges.
