Machine Learning Material Search for Fast Candidate Screening

The patent applies preliminary action by performing HTS and machine learning evaluations before conducting full simulations. Virtual materials are pre-screened using computationally inexpensive methods to identify promising candidates, so that only a small subset undergoes expensive simulation. This resolves the contradiction by doing the necessary selection work in advance, avoiding the need to simulate all possible materials.

The patent segments the material discovery process into multiple stages: initial HTS generation, machine learning-based property prediction, and targeted simulation of top candidates. This segmentation allows the system to handle large numbers of virtual materials efficiently by processing them in groups with increasing computational intensity, rather than simulating all materials at full computational cost.

The patent merges multiple computational methods into a hybrid workflow: HTS generates virtual materials, machine learning models predict properties quickly, and simulation provides detailed validation. This combination allows the system to maintain high productivity through ML while ensuring precision through selective simulation of top candidates, resolving the contradiction between speed and accuracy.

The patent implements feedback by using simulation results to train and refine machine learning models. The ML models learn from accurate simulation data to improve their predictions, creating a feedback loop that enhances selection accuracy over time while maintaining high productivity. This resolves the contradiction by making the ML predictions progressively more reliable.