Project Summary In order to translate multi-modal sensory information into goal-directed behavior, both brains and artificial neural networks construct abstract internal representations. The nature, mechanistic origin, and function of these internal representations are currently a subject of high interest. Here we propose to investigate two circuits in the central brain of Drosophila the allow multi-modal sensory information to be translated into goal- directed navigation. First, we will examine the integration of optic flow, airflow, and locomotor information in the columnar input pathway to the fan-shaped body, a part of the Drosophila navigation system. We hypothesize that this circuit allows the fly to adapt to altered sensory input and to disambiguate sensation caused by its own movement from that caused by external stimuli. Second, we will investigate the dynamics of chemo-sensory and self-motion representations in local neurons of the fan-shaped body. We will test the connectome-driven hypothesis that recurrent excitation and global inhibition interact to shape the distinct spatio-temporal dynamics of representations in different neuron types. Finally, we will use computational approaches to investigate the function of these representations. We will construct circuit models based on experimental data to test the proposed function of each circuit and also examine the conditions under which artificial neural network trained to perform ethological navigation tasks develop similar internal representations. Together these experimental and computational investigations will yield insight into the nature, circuit origins, and function of multi-modal internal representations.