In this talk, I will discuss how the rigorous application of machine learning techniques on large materials data sets may be used to develop efficient, quantum-accurate force-fields for elemental metals (e.g., Mo, Li) as well as complex multi-species compounds. We will outline a systematic approach to structural selection based on principal component analysis, as well as a novel differential evolution algorithm for optimizing the hyperparameters in the model fitting so that both the model error and the property prediction error can be simultaneously lowered. We will demonstrate that this force field can successfully predict a broad range of properties, such as energies, forces, elastic constants, melting point, phonon spectra, surface energies, etc. with accuracy close to that of DFT computations, outperforming traditional force fields based on the embedded atom (EAM) and modified embedded atom methods (MEAM). We expect that these techniques will find broad application in large-scale, long-time scale simulations.