Machine learning meets the power of experts
In our article, we explore how machine learning can assist experts in conceiving new microstructural features that help improve the accuracy of learned microstructure-property realtionships. In contrast to integrating the human in the loop, another approach that builds on convolutional neural networks is proposed. We develop an accurate and robust network based on our novel deep inception module. Interestingly, it not only outperforms established frameworks such as ResNet but also has fewer parameters to optimize.
The first version of the article (prior to the revision) is available as a preprint (see below). However, we hope the final manuscript is soon to be published - watch this space!
Update (Oct 14, 2024): DOI of the article is 10.1186/s40323-024-00275-1 (opens in new window; not accessible as of today - production of the article is pending)
Preprint at Research Square (original submission, before the revision)