Mammals show a significant capacity for contextual knowledge transfer, applying acquired information to novel situations. Such generalization involves extracting common features from diverse experiences and utilizing them to make predictions in unfamiliar contexts. It remains unknown which brain areas are responsible for such abstractions. Neural representations in the medial entorhinal cortex (MEC) constitute a generalized map of the local space, but recent findings may suggest that the MEC possesses non-spatial dimensions suggesting a generalized network for learning state spaces irrespective of modalities. In this study, we investigate the functional properties of the dorsal layer II/III MEC neurons during learning of a non-spatial, go/no-go visual association task. Head-fixed mice are trained over multiple sessions while recording neural activity using two-photon imaging. Utilizing our newly developed soma-targeted calcium sensor (Soma-GCaMP8s) we have recorded the activity of identified neurons from the animal's initial naive state to an expert level performance in the behavioral task. To examine the generalization of structural components within the task, we introduce rule changes while monitoring performance and neural responses during learning the new rules. We find learning-induced changes in the MEC population, with an increasing fraction of the population showing tuning to the rewarded visual cue. Strikingly, following the same neuron population over the course of learning, shows that the development of the response pattern is reflected in discrete anatomical cell clusters in the field of view. Upon rule switching, a fraction of the population follows the cue-outcome while a small number of units follows the cue identity.