Biologically inspired recurrent neural network modeling of an evidence integrator circuit.

Evidence accumulation is a fundamental neural computation essential for adaptive behavior, yet its synaptic implementation remains unclear. Addressing this challenge critically depends on linking neural dynamics to circuit structure within the same brain. Here, we combine functional calcium imaging with large-scale ultrastructural electron microscopy (EM) to uncover the wiring logic of visual evidence accumulation in larval zebrafish. In a functionally imaged EM dataset of the anterior hindbrain, we identify conserved morphological cell types whose activity patterns define distinct computational roles. Bilateral inhibition, disinhibition, and sparse recurrent connectivity emerge as key circuit motifs shaping these. Based on these results, we train a connectome-constrained neural network model using observed dynamics that yields predictions we tested and confirmed experimentally. Our work illustrates how a combination of several connectivity motifs implements bilateral sensory evidence accumulation in a vertebrate brain.

My research explores how nervous systems transform sensory information into decisions and actions. Using larval zebrafish as a model system, my lab investigates the neural algorithms and circuit mechanisms that allow animals to evaluate their environment, integrate evidence over time, and select adaptive behaviors. A central goal of our work is to understand how relatively small neural circuits can implement sophisticated computations underlying perception, decision-making, and intelligence. To address these questions, we combine behavioral experiments, virtual reality technologies for freely behaving animals, and large-scale neural recording and manipulation. These approaches allow us to link sensory stimuli, neural activity, and behavior with high precision. In parallel, we use advanced microscopy and molecular techniques to uncover the cellular and circuit architectures that implement these computations in the brain. More broadly, our research aims to bridge levels of analysis—from genes and synapses to neural circuits, algorithms, and collective behavior. By integrating experimental neuroscience with computational modeling and emerging molecular methods, we seek to uncover general principles of how brains process information and generate intelligent behavior, both in individuals and in interacting animal groups.