We propose a Deep Learning-based calibration algorithm for material model parameters for rocks simulated using Discrete Element Method (DEM). We considered granite and chalk as representatives of rocks of different geological origins. We create Discrete Element models using Bonded-Particle Method (BPM) to simulate Uniaxial Compression tests and Brazilian tests commonly used for experimental determination of material properties. After initial simulations, the microscopic parameters of the model are fed into a constructed DL system based on the Multi-Layer Perceptron regressor as input data and the functional depending on macroscopic parameters as the output. Additional optimization algorithm called Dual Annealing was later used to find the local minimum of the functional and to identify the optimal parameter set. As a result, a set of optimal microscopic parameters has been found and tested in a simulation using the DEM model of the same size. Afterwards, we compared the resulting stress–strain curve with the experimental data used as a reference for calibration. Additionally, the stability of the obtained model in terms of reproducibility of the stress–strain curve has been tested by simulating Uniaxial Compression with different particle packing. The discrete element models obtained with material parameters show stability and consistency with experimental data. We conducted a correlation analysis, from which we were able to see the dependence of target parameters (critical stresses) on the variation of the calibrated parameters. The analysis shows that bond diameter is the most important parameter, whose calibration should be prioritized. Statistical indicators that we computed have only reinforced that conclusion. This procedure can be used to simulate rocks in geomechanical and geoengineering applications requiring discontinuous description of rock behavior, such as multiple crack formation, drilling, boring, and rock blasting.