![Climate model of Europe: 2003 air temperature
Daily fluctuations in the air temperature
[◦
C]
Tensor: (424 × 412 × 365)
(columns × rows × days)
NTFk applied to extract dominant
hidden (latent) features
ML Studies Climate: EU Summary](https://image.slidesharecdn.com/isr2019-190516201930/85/Physics-Informed-Machine-Learning-Methods-for-Data-Analytics-and-Model-Diagnostics-10-320.jpg)
![Climate model of Europe: 2003 water-table depth
Daily fluctuations in the water-table
depth [m]
Tensor: (424 × 412 × 365)
(columns × rows × days)
NTFk applied to extract dominant
hidden (latent) features
ML Studies Climate: EU Summary](https://image.slidesharecdn.com/isr2019-190516201930/85/Physics-Informed-Machine-Learning-Methods-for-Data-Analytics-and-Model-Diagnostics-13-320.jpg)
![Climate model of Europe: air temperature 1989-2017
Monthly fluctuations in the air
temperature from 1989 to 2017 [◦
C]
Tensor: (316 × 316 × 348)
(columns × rows × months)
NTFk applied to extract dominant
hidden (latent) features
ML Studies Climate: EU Summary](https://image.slidesharecdn.com/isr2019-190516201930/85/Physics-Informed-Machine-Learning-Methods-for-Data-Analytics-and-Model-Diagnostics-20-320.jpg)
Physics-Informed Machine Learning Methods for Data Analytics and Model Diagnostics
- 1. Physics-Informed Machine Learning Methods for Data Analytics and Model Diagnostics Velimir V. Vesselinov (monty) (vvv@lanl.gov) Earth and Environmental Sciences Division, Los Alamos National Laboratory, NM, USA http://tensors.lanl.gov LA-UR-19-24562 ML Studies Climate: EU Summary
- 2. Machine Learning (ML) We have developed a series of novel Machine Learning (ML) methods Data Analytics: Identify features (signals) in datasets Model Analytics: Identify processes (features) in model outputs Physics-Informed Machine Learning: Coupled Data/Model Analytics ML Studies Climate: EU Summary
- 3. Physics-Informed Machine Learning (PIML) Extracting common hidden features present in data (observations/experiments) and model outputs Incorporating physics constraints in ML analyses Incorporating physics processes in ML analyses ML Studies Climate: EU Summary
- 4. Physics-Informed Machine Learning (PIML) PIML is of critical importance for data- and physics-driven science and engineering applications PIML models can replace computationally intensive numerical simulators PIML models can be applied to predict complex processes when: • physics is not well understood • numerical models cannot be developed (due to physics complexity) Training PIML models is 1-2 orders of magnitude faster than traditional ML models ML Studies Climate: EU Summary
- 5. Physics-Informed Machine Learning (PIML): Cartpole “Deep Learning is dead. Long Live Differentiable Programming!” Yann LeCun, Chief AI Scientist at Facebook, and one of the leading ML experts in the world Differentiable Programming for PIML can be done efficiently only in C/C++, FORTRAN, Python, scikit-learn, TensorFlow, PyTorch cannot do it ML Studies Climate: EU Summary
- 6. Physics-Informed Machine Learning (PIML): Cartpole ML Studies Climate: EU Summary
- 7. ML Analyses Field Data: Groundwater contamination US Climate data Geothermal data Seismic data Lab Data: X-ray Spectroscopy UV Fluorescence Spectroscopy Microbial population analyses Operational Data: LANSCE: Los Alamos Neutron Accelerator Oil/gas production Model Outputs: Reactive mixing A + B → C Phase separation of co-polymers Molecular Dynamics of proteins EU Climate modeling ML Studies Climate: EU Summary
- 8. ML Publications Vesselinov, Munuduru, Karra, O’Maley, Alexandrov, Unsupervised Machine Learning Based on Non-Negative Tensor Factorization for Analyzing Reactive-Mixing, Journal of Computational Physics, (in review), 2019. Stanev, Vesselinov, Kusne, Antoszewski, Takeuchi, Alexandrov, Unsupervised Phase Mapping of X-ray Diffraction Data by Nonnegative Matrix Factorization Integrated with Custom Clustering, Nature Computational Materials, 2018. Vesselinov, O’Malley, Alexandrov, Nonnegative Tensor Factorization for Contaminant Source Identification, Journal of Contaminant Hydrology, 2018. O’Malley, Vesselinov, Alexandrov, Alexandrov, Nonnegative/binary matrix factorization with a D-Wave quantum annealer, PLOS ONE, 2018. Vesselinov, O’Malley, Alexandrov, Contaminant source identification using semi-supervised machine learning, Journal of Contaminant Hydrology, 10.1016/j.jconhyd.2017.11.002, 2017. Alexandrov, Vesselinov, Blind source separation for groundwater level analysis based on nonnegative matrix factorization, WRR, 10.1002/2013WR015037, 2014. ML Studies Climate: EU Summary
- 9. Climate model of Europe TerrSysMP: coupled climate simulator (CLM/ParFlow) of atmospheric, surface and subsurface processes ML Studies Climate: EU Summary
- 10. Climate model of Europe: 2003 air temperature Daily fluctuations in the air temperature [◦ C] Tensor: (424 × 412 × 365) (columns × rows × days) NTFk applied to extract dominant hidden (latent) features ML Studies Climate: EU Summary
- 11. Climate model of Europe: 2003 temperature fluctuations represented by 3 features ML Studies Climate: EU Summary
- 12. Climate model of Europe: 2003 temperature fluctuations represented by 3 features ML Studies Climate: EU Summary
- 13. Climate model of Europe: 2003 water-table depth Daily fluctuations in the water-table depth [m] Tensor: (424 × 412 × 365) (columns × rows × days) NTFk applied to extract dominant hidden (latent) features ML Studies Climate: EU Summary
- 14. Climate model of Europe: 2003 water-table fluctuations represented by 3 features ML Studies Climate: EU Summary
- 15. Climate model of Europe: Water-table fluctuations represented by 2 features ML Studies Climate: EU Summary
- 16. Climate model of Europe: Water-table fluctuations represented by 3 features ML Studies Climate: EU Summary
- 17. Climate model of Europe: Water-table fluctuations represented by 4 features ML Studies Climate: EU Summary
- 18. Climate model of Europe: Water-table fluctuations represented by 5 features ML Studies Climate: EU Summary
- 19. Climate model of Europe: Water-table fluctuations represented by 6 features ML Studies Climate: EU Summary
- 20. Climate model of Europe: air temperature 1989-2017 Monthly fluctuations in the air temperature from 1989 to 2017 [◦ C] Tensor: (316 × 316 × 348) (columns × rows × months) NTFk applied to extract dominant hidden (latent) features ML Studies Climate: EU Summary
- 21. Climate model of Europe: 2003 air temperature reconstruction by 3 features Original Reconstruction Error ML Studies Climate: EU Summary
- 22. Climate model of Europe: 2003 air temperature reconstruction by 4 features Original Reconstruction Error ML Studies Climate: EU Summary
- 23. Climate model of Europe: 2003 air temperature reconstruction by 5 features Original Reconstruction Error ML Studies Climate: EU Summary
- 24. Climate model of Europe: 2003 air temperature reconstruction by 6 features Original Reconstruction Error ML Studies Climate: EU Summary
- 25. Climate model of Europe: 2003 air temperature reconstruction by 7 features Original Reconstruction Error ML Studies Climate: EU Summary
- 26. Climate model of Europe: 2003 air temperature reconstruction by 8 features Original Reconstruction Error ML Studies Climate: EU Summary
- 27. Climate model of Europe: 2003 air temperature reconstruction by 9 features Original Reconstruction Error ML Studies Climate: EU Summary
- 28. Climate model of Europe: 2003 air temperature reconstruction by 10 features Original Reconstruction Error ML Studies Climate: EU Summary
- 29. Climate model of Europe: 2003 air temperature reconstruction by 15 features Original Reconstruction Error ML Studies Climate: EU Summary
- 30. Climate model of Europe: 2003 air temperature reconstruction by 20 features Original Reconstruction Error ML Studies Climate: EU Summary
- 31. Climate model of Europe: 2003 air temperature reconstruction by 25 features Original Reconstruction Error ML Studies Climate: EU Summary
- 32. Climate model of Europe: 2003 air temperature reconstruction by 30 features Original Reconstruction Error ML Studies Climate: EU Summary
- 33. Climate model of Europe: 2003 air temperature reconstruction by 35 features Original Reconstruction Error ML Studies Climate: EU Summary
- 34. Climate model of Europe: 2003 air temperature reconstruction by 40 features Original Reconstruction Error ML Studies Climate: EU Summary
- 35. Climate model of Europe: 2003 air temperature reconstruction by 45 features Original Reconstruction Error ML Studies Climate: EU Summary
- 36. Climate model of Europe: 2003 air temperature reconstruction by 50 features Original Reconstruction Error ML Studies Climate: EU Summary
- 37. Climate model of Europe: air temperature reconstruction errors ML Studies Climate: EU Summary
- 38. Climate model of Europe: air temperature features (8) ML Studies Climate: EU Summary
- 39. Climate model of Europe: air temperature features (8) 1989-2017 ML Studies Climate: EU Summary
- 40. Climate model of Europe: air temperature features (8) 2002-2003 ML Studies Climate: EU Summary
- 41. Climate model of Europe: air temperature reconstruction by 8 features Original Reconstruction Error ML Studies Climate: EU Summary
- 42. Climate model of Europe: air temperature reconstruction by 50 features Original Reconstruction Error ML Studies Climate: EU Summary
- 43. Climate model of Europe: analyze all model outputs (>40) simultaneously Water-table depth Temperature Evaporation Sensible heat flux Find interconnections and common dominant features in model outputs Evaluate impacts of different model setups (initial conditions, resolution, etc.) Find dominant processes impacting model predictions (e.g., occurrence of heat waves, climate impacts on groundwater resources, climate impacts of subsurface processes on atmospheric conditions) ML Studies Climate: EU Summary
- 44. Summary Developed a series of novel ML methods and computational tools Our ML methods have been applied to solve various real-world problems at various scales (from molecular to global) (brought breakthrough discoveries related to human cancer research) Modeldata Sensor data Experimentaldata Robust Unsupervised Machine Learning ML Studies Climate: EU Summary
- 45. Machine Learning (ML) Algorithms / Codes developed by our team Codes: NMFk.jl Mads.jl NTFk.jl Examples: http://madsjulia.github.io/Mads.jl/Examples/blind_source_separation http://tensors.lanl.gov http://tensordecompositions.github.io ML Studies Climate: EU Summary