In brain networks, neurons communicate through action potentials or spikes. These spikes can be thought of as discrete events that constitute, in essence, a binary, time-varying spatial pattern over the entire network. A general hypothesis in neuroscience is that these patterns encode information, thus enabling function. Consequently, an emerging research direction in experimental neuroscience involves the use of neurostimulation technologies to artificially induce such patterns in a spatiotemporally precise manner - the so-called neurocontrol problem. In this work, we discuss the neurocontrol problem by means of statistical models, which, in contrast to more traditional dynamical-systems models, describe only the probability of spiking as a function of time. Thus, such models aggregate nonlinearity and uncertainty into a more tractable mathematical description. While statistical models are frequently used to describe experimental data, their use as tools for input construction is not as well explored. Here, we formulate an optimal control problem for spiking patterns via a weighted maximum likelihood approach and develop its solution. We demonstrate the design approach for a model network consisting of coupled stochastic integrate-and-fire neurons. Finally, we suggest how this overall framework can be used to develop a class of control analyses for point process models.