The understanding of dislocation microstructures and their evolution is crucial for predicting the mechanical behaviour of crystalline materials. While continuum theories and computational models describe dislocation microstructures by dislocation densities and related field variables, the resolution at the micro-scale is limited, where the discrete nature of dislocations governs key material properties. Resolving the dislocation microstructure on the micro-scale typically requires computationally expensive molecular dynamics (MD) or discrete dislocation dynamics (DDD) simulations. To connect dislocation microstructures in the micro-scale with those in the meso-scale represented by continuum quantities, generative machine learning can be employed to learn the relationship between continuum inputs and microstructural details 

This contribution introduces a generative machine learning approach to connect discrete dislocation microstructures with continuum field variables. The generative models are trained on datasets representing three-dimensional dislocation microstructures from discrete dislocation dynamics simulations. This contribution demonstrates that generative models can produce statistically and physically plausible dislocation configurations with analogous effective properties conditioned on continuum inputs such as dislocation density, geometrically necessary dislocations (GND) density, and additional graph features. The reconstructions capture essential features like dislocation network connectivity and formation, enabling high-fidelity microscale interpretations from continuum data. This approach extends coarse graining methods opening new pathways for data-driven multiscale modelling in crystal plasticity.