Abstract One of the major barriers to understanding how neural circuits give rise to behavior is that typical experimental preparations make it difficult to study these circuits across different brain areas. Recent advances in microscopy and calcium sensors have made it possible to simultaneously record up to thousands of individual neurons, and optical methods have made it possible to stimulate hundreds at a time, but current approaches, which stimulate only subsets of predetermined neurons, are not adequate for dissecting large-scale neural circuits. Here, we propose to develop a novel integrated experimental- computational platform to test neural circuit hypotheses of the zebrafish optomotor response, a representative sensorimotor behavior. This platform will allow us to characterize the relationships among functionally defined groups of neurons as the data are collected in real-time. By using prior-guided algorithms that adaptively choose scanless 3D holographic photostimulation patterns of up to hundreds of neurons in response to previously observed data, we will be able to exponentially increase data efficiency, simultaneously inferring multiple classes of functional connections between visually responsive neurons in the zebrafish pretectum and their downstream targets. Once established, this approach will allow us to perform adaptive experiments that selectively perturb neural function based on function, accelerating the process of model generation and hypothesis testing. Moreover, these tools will be applicable to other types of calcium imaging data, with broad implications for systems neuroscience.