Kolmogorov-Arnold Networks for High-Entropy Alloys Design

Deep learning-based machine learning techniques have been extensively applied to the design of high-entropy alloys (HEAs), yielding numerous valuable insights. However, challenges remain in accurately predicting outcomes with sparse and complex datasets and ensuring the interpretability of the models. Kolmogorov-Arnold Networks (KANs) are a recently developed architecture that aims to improve both the accuracy and interpretability of input features. In this work, we explored three different datasets for HEA design and demonstrated the application of KANs for both classification and regression models. In the first example, we used a KAN classifier model to predict the probability of single-phase formation in High Entropy Carbide Ceramics (HECCs) based on various physical properties. In the second example, we employed a KAN regressor model to predict the yield strength and ultimate tensile strength of HEAs based on their chemical composition and mechanical properties. The third example involved a classifier model to determine whether a certain elemental composition is an HEA or non-HEA, followed by a KAN regressor model to predict the bulk modulus of the identified HEA, aiming to identify HEAs with high bulk modulus. In all three examples, KANs either outperformed or matched the performance of existing machine learning models, particularly the Multilayer Perceptron (MLP), which served as our point of comparison. Our work opens new opportunities for adopting novel AI algorithms in the design and development of HEAs. By demonstrating the efficacy of KANs in handling both classification and regression tasks, we provide a promising direction for future research to explore advanced machine learning techniques, which lead to more accurate predictions and better interpretability of complex materials, ultimately accelerating the discovery and optimization of HEAs with desirable properties.