Abstract: In crystalline materials, dislocations define the plastic response under most conditions. Thus, the method of discrete dislocation dynamics (DDD), introduced as a coarse-grained model of true atomistic dynamics of dislocations, has been widely employed to investigate crystal plasticity at the mesoscale. However, its applicability remains limited by its computational cost and physical fidelity. In this talk, we will present the recent advances we have made to address both of these limitations using graph neural networks (GNN) to learn dislocation dynamics. First, we will introduce the DDD-GNN model [1] in which we propose to substitute the expensive time-integration procedure of dislocation motion with a GNN model trained on DDD trajectories. Second, we will present the DDD+ML model [2], a data-driven framework in which the conventional dislocation mobility function is replaced by a GNN model trained on large-scale MD trajectories, thereby providing a novel approach for calibrating and improving prediction fidelity of the DDD method. [1] N. Bertin, F. Zhou, Accelerating discrete dislocation dynamics simulations with graph neural networks, Journal of Computational Physics 487, 112180 (2023) [2] N. Bertin, V. Bulatov, F. Zhou, Learning dislocation dynamics mobility laws from large-scale MD simulations, arXiv preprint arXiv:2309.14450 (2023) This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344."