Magnetic Resonance Imaging (MRI) can be accelerated by sampling below the Shannon-Nyquist rate via compressed sensing techniques. In this talk, I will consider the problem of optimizing the under-sampling pattern in a data-driven fashion, which has been an open problem for over a decade. For a given sparsity constraint, our method optimizes the under-sampling pattern and reconstruction model, using a computationally efficient end-to-end deep-learning strategy. We call our method Learning-based Optimization of the Under-sampling PattErn, or LOUPE. Our experiments with brain and knee MRI scans show that the LOUPE-derived pattern can yield significantly more accurate reconstructions compared to standard under-sampling schemes. I will also present results for prospectively collected in-vivo images that demonstrate the practical utility of LOUPE in speeding up MRI scans.