My research focuses on understanding how biologically realistic networks carry out computations. More specifically, my goal is to understand how connectivity and single neuron properties determine the ability of networks to perform computations, and how synaptic learning rules, single neuron properties, and patterns of activity determine connectivity. This leads to four main questions: How do networks extract only the relevant information from the outside world and transform it into a representation that is efficient for computations carried out by downstream neurons? How do networks retain uncertainty, or even whole probability distributions? What aspects of neural activity carries information? And how can networks learn to solve these problems? To answer these, I combine analytical approaches -- taken mainly from dynamical systems and statistical physics -- with large scale simulations of spiking networks.