Complex brain functions are realized through the coordinated activity of populations of neurons. To manipulate brain function, it is crucial for brain stimulation techniques to achieve targeted control at the level of neural populations. Electrical microstimulation (uStim) has the potential to create targeted activity states because different combinations of stimulation parameters (e.g., location of the stimulated electrode(s), current amplitude, and frequency) produce diverse neural population activity states. However, its use for this purpose has been hindered by the challenge of identifying optimal stimulation parameters among the large number of possible combinations. To address this problem, we developed MicroStimulation Optimization (MiSO), a closed-loop stimulation framework to drive neural population activity toward specified states by optimizing over a large stimulation parameter space. MiSO leverages three key methodologies: 1) performing latent space alignment to statistically merge stimulation-response samples across sessions, 2) using a statistical model trained on the merged dataset to predict the brain’s response to unseen stimulation parameters, and 3) integrating the model’s prediction into an online parameter update algorithm to adaptively change stimulation parameters within a session. We tested MiSO in closed-loop experiments using uStim in the prefrontal cortex of a non-human primate. Guided by the model predictions, MiSO successfully searched amongst thousands of stimulation parameter configurations to drive the neural population activity toward specified states. More broadly, MiSO increases the clinical viability of neuromodulation technologies by enabling the use of many-fold larger stimulation spaces.