Tag: Magneto-Ionics

For machine learning (ML) in high-tech materials development, where high-quality experimental datasets are typically small (hundreds of samples), developing reliable models requires specialized techniques like Uncertainty Quantification (UQ). UQ focuses on making a model "know what it doesn't know" by requiring it to predict not just the material property, but also its confidence level in that prediction. A common approach to achieve this is ensembling, where multiple models are trained slightly differently, and the standard deviation of their individual predictions is used as the measure of uncertainty. This uncertainty is critical because it allows for strategic post-processing, such as filtering out highly uncertain predictions (a process known as sparsification) to significantly reduce the overall prediction error, thereby enabling high-accuracy predictive modeling of new materials despite the constraints of limited data.