Transfer Learning for Multi-material Classification of Transition Metal Dichalcogenides with Atomic Force Microscopy

Deep learning models are widely used for the data-driven design of materials based on atomic force microscopy (AFM) and other scanning probe microscopy. These tools enhance efficiency in inverse design and characterization of materials. However, limited and imbalanced experimental materials data typically available is a major challenge. Also important is the need to interpret trained models, which have typically been complex enough to be uninterpretable by humans. Here, we present a systemic evaluation of transfer learning strategies to accommodate low-data scenarios in materials synthesis and a model latent feature analysis to draw connections to the human-interpretable characteristics of the samples. Our models show accurate predictions in five classes of transition metal dichalcogenides (TMDs) (MoS\(_2\), WS\(_2\), WSe\(_2\), MoSe\(_2\), and Mo-WSe\(_2\)) with up to 89\(\%\) accuracy on held-out test samples. Analysis of the latent features reveals a correlation with physical characteristics such as grain density, DoG blob, and local variation. The transfer learning optimization modality and the exploration of the correlation between the latent and physical features provide important frameworks that can be applied to other classes of materials beyond TMDs to enhance the models' performance and explainability which can accelerate the inverse design of materials for technological applications.