EXTENDED ABSTRACT: In the field of materials science, the majority of tasks we encounter are inherently multimodal, involving distinct data modalities such as crystal structures, molecular structures, characterization spectra, structure-activity relationships, and service performance. Consequently, multimodal machine learning approaches are becoming increasingly important in the machine learning applications within materials science. Contrastive learning, as a more mature bidirectional bridge across different modalities in multimodal tasks, undoubtedly deserves attention in materials science as well. The success of the CLIP model as the backbone of the DALL-E 2 model has demonstrated the potential of this approach in crossmodal representation learning. In recent years, contrastive learning has gained widespread adoption in machine learning applications to physical systems, primarily due to its distinctive crossmodal capabilities and scalability. Building on the foundation of Kolmogorov-Arnold Networks (KANs[1]), we introduce a novel contrastive learning framework, Kolmogorov-Arnold Contrastive
Crystal Property Pre-training (KCCP), which integrates the principles of CLIP and KAN to establish robust correlations between crystal structures and their physical properties. During the training process, we conducted a comparative analysis between Multi-Layer Perceptron (MLP) and KAN, revealing that KAN significantly outperforms MLP in both accuracy and convergence speed for this task. By extending the capabilities of contrastive learning to the realm of physical systems, KCCP offers a promising approach for constructing cross-data structural and cross-modal physical models, representing an area of considerable potential. 

Keywords：Contrastive learning; Kolmogorov-Arnold Network; Physics Laws
REFERENCES:
[1] Liu, Ziming, et al. "Kan: Kolmogorov-arnold networks." arXiv preprint arXiv:2404.19756 (2024).