Data Driven Discovery

Data-driven discovery uses machine learning to automate the process of scientific hypothesis generation and verification directly from datasets, aiming to accelerate scientific progress and reduce reliance on traditional experimental methods. Current research focuses on developing and benchmarking algorithms like Physics-Informed Neural Networks (PINNs), sparse identification of nonlinear dynamics (SINDy), and various deep learning architectures (including graph neural networks and generative models) to identify governing equations, discover self-similarity, and extract meaningful insights from complex data. This approach holds significant promise for diverse fields, enabling faster materials discovery, improved climate modeling, and more accurate predictions in areas like fluid dynamics and rogue wave forecasting, ultimately enhancing scientific understanding and technological innovation.