Conference on Cognitive Computational Neuroscience

Humans are capable of generalizing using the similarity between the abstract features of external stimuli. The neural mechanisms underlying generalization using abstract representations are unclear, and future discoveries would be facilitated by studies in animal models. Here we developed a rat model of generalization by leveraging an existing pulse-based evidence accumulation task, with the abstract rule of choosing the side with the greater number of light pulses. We trained rats (n=14) on different curricula that limit the stimulus-action associations presented during learning and then used test trials to evaluate generalization to novel stimuli. The performance of rats was consistent with a hybrid Exemplar-to-Prototype model of learning. In this model, sensory evidence in the training trials is altered by internal noise providing an expanded perceptual experience to guide generalization. Overall, our study suggests that pulse-based accumulation tasks may be used to study the neural mechanisms of generalization in animals and highlights how noise may support cognition.