Scientific ML/AI
Shall we call it "Scientific ML" or "Physics-informed machine learning?" I think PINN is a subset, so let's stick with SciML.
Project notes
we'll start with examples.
Common SciML approaches to solving problems
Surrogates
Totally take out the physics with a NN approximation.
Domains:
- EDA
- MD
- Fluids
Inverse
- MD
- Fluids
ML-driven
- EDA
- "faster monte-carlo"
Hybrid
- Fluids
Other?
- is there a better term for our "Physics-informed DRL?" I.e., this is bigger and more complex than PINN but also perhaps closer to the problem domain?
Worked examples
WIP
Target audience
Potential learners:
- Researchers: know the science, new to deep learning
- Data Scientists: know the ML, new to the science
- Undergrads: new to both
References
-
some motivation from our fearless leader
-
a seismic example
-
Learning Particle Physics by Example: Location-Aware Generative Adversarial Networks for Physics Synthesis