For the olfactory associative learning to be specific, stimulus representation in the Kenyon cell (KC) population should be accurate with minimal activity overlap between different stimuli, which is achieved by the sparse and highly odor-specific KC response patterns. While a multitude of cellular and network mechanisms substantially increase the KC population sparsity, the proportion of KCs responding unreliably to a given odor over multiple trials is roughly three times larger than that of reliable responders [1, 3], leading to substantial response correlations between different odors. Considering dopamine signaling in the mushroom body lobes is unspecific with volume transmission, population sparseness alone is not sufficient to ensure specificity in associative learning. It has been recently shown, that additional axo-axonic interactions with muscarinic type-B receptors (mAChR-B) between KCs enhance the specificity in associative learning by suppressing both odor mediated Ca2+ signals and dopaminergic neuron driven cAMP signals [2], increasing learning specificity by inhibiting KCs responding unreliably to a given odor. In this project, we theoretically investigate the functions and significance of lateral inhibition in learning specificity by comparing variants of KC population rate models with and without lateral inhibition. In line with the experimental observations, the naive model without lateral inhibition shows a lower degree of learning specificity when compared to the full model. Finally, we address possible mAChR-B related molecular mechanisms, such as activity-dependent gating of lateral inhibition, which leads to an overall enhancement in learning performance. Ultimately, we aim to propose an extended model for local mushroom body bouton computation, highlighting the importance of lateral interactions between KCs in learning and memory [4].