Chemical Machine Learning TRIZ Cases
- Weighted Masking for Molecular Graph Neural Network Pre-Training
Frequency-based atom masking gives rare elements more pre-training exposure, improving molecular property prediction without added model modules.
- Generative ML Inversion for Borehole Sensing
Generative machine learning models map borehole measurements to material properties, replacing computationally expensive traditional simulation methods.