New techniques in neuroscience have opened the door to rich new data sets of neural activities. These data sets shed light on the computational foundation of the brain, i.e., neurons and synapses. However, these data also present unprecedented challenges: novel statistical theory and methods are required to model neural activities, and well-designed experiments are needed to collect informative data. In this talk, we take on the task of learning connectivity among large sets of neurons. In particular, we discuss i) how to learn functional connectivity from spike train data using the Hawkes process, and ii) how to optimally design experiments to collect data that allow us, for the first time, to learn physiological connectivity in vivo.