Recently, machine learning (ML) has emerged as an additional scientific methodological approach besides theory, experiment, and computation. Although not a new concept, ML has emerged as a powerful tool in materials science, particularly for predicting material properties based on chemical composition. However, it is very common for some computer-aided applications, such as those in computational sciences and ML, to utilize ”black box” approaches, which sacrifice interpretability and may lead to computational artifacts, and erroneous conclusions. This is why, we plan to implement physics-guided ML (PGML), where interpretability is maintained to have new insights and where each feature should hold physical significance. Through this seminar, I present the implementation of PGML for predicting materials properties from chemical composition. Additionally, this will enhance the understanding of relationships between chemical composition and material properties, leading the way for more efficient material design and discovery processes.

All faculty, researchers and students are invited to attend.