| Abstract Scope |
Material discoveries for improved societal, environmental, safety device applications etc. often require optimization of expensive experimental process for materials synthesis, characterization, and learning structure-property; or require analysis of large complex experimental data, such as EELS or 4D STEM, that contain information on structural, physical, and chemical properties of materials. This resulted in strong interest in applying data-driven methods to accelerate learning. However, such classical data-driven approaches do not always guarantee to provide alignment to the domain scientist intended learning objectives. To improve such alignment, I will showcase a few examples of combining existing physics and on-the-fly gained domain knowledge into data-driven methods. Here, the physical information, injected through minimal or no human intervention, allows us to better steer towards discovery, while the ML policy allows to accelerate the discovery. The proposed approach highlights the improvement towards meaningful scientific discoveries in material research, with potential opportunities to considerably more complex problems. |