Applications of Physics-Informed Neural Networks in Science: A Hands-On Tutorial
Saturday, April 18, 2026 · 9:00 AM – 10:30 AM · Classroom 206
Scientific problems face a paradox: we have centuries of physics knowledge, yet modern deep learning often ignores it. Physics-Informed Neural Networks (PINNs) solve this by embedding physical laws—encoded as differential equations—directly into neural network training. The result? Models that learn from both data and physics, requiring less labeled data while respecting known constraints.
This 90-minute hands-on tutorial takes you from theory to working code. Every concept is paired with a ready-to-run Google Colab notebook—no installation required.
What You’ll Build:
1) Neural Network 101- How does NN and hidden layers work? 2) Harmonic Oscillator — Your first PINN: solving ODEs with neural networks 3) How to use PINNs to fill in your climate data gaps. 4) Hyperelastic Material Deformation — Nonlinear solid mechanics without finite element meshes, comparing Feed-Forward Networks vs. Kolmogorov-Arnold Networks (KAN)
Key Takeaways:
✓ How PDEs become loss functions via automatic differentiation ✓ The hybrid Adam → L-BFGS optimization strategy critical for PINN convergence ✓ When PINNs outperform traditional numerical methods (and when they don’t) ✓ Practical debugging: failure modes and adaptive loss weighting ✓ 4+ production-tested Colab notebooks to take home Who Should Attend: ML engineers exploring scientific applications, computational scientists seeking mesh-free alternatives, and domain experts wanting to incorporate physics into ML pipelines. Basic Python and neural network familiarity assumed; no physics background required.
All materials are open-source and have been refined through university courses and research workshops reaching hundreds of students and practitioners.
Textbook: https://leanpub.com/generativeaiforscience/c/AMLC26
About the Speaker
Paul Liu
Professor at NC State University
Dr. J. Paul Liu is a Professor at NC State University (Department of Marine, Earth, and Atmospheric Sciences), where he teaches "Generative AI for Science" through the Data Science and AI Academy. Expertise: Physics-Informed Neural Networks for coastal forecasting, Graph Neural Networks for water quality prediction, and AI applications in climate science. Publications: Generative AI for Science (2026) — 540-page guide with 50+ Colab notebooks covering AI for chemistry, biology, physics, and climate How to Build and Fine-Tune a Small Language Model (2025) (Leanpub/Amazon, 479 pages) Teaching Experience: University courses and research workshops for students across geoscience, engineering, and data science—from complete beginners to advanced practitioners. Research: work on multi-architecture physics-informed deep learning for coastal ocean forecasting (collaborative project with Louisiana State University). Dr. Liu brings both theoretical depth and production experience deploying PINNs for real-world scientific problems, ensuring attendees receive battle-tested techniques that work beyond toy examples.