Repetitive activation of subpopulations of neurons leads to the formation of neuronal assemblies, which can guide learning and behavior. Recent technological advances have made the artificial induction of such assemblies feasible, yet how various parameters of perturbation can be optimized for such induction is not clear. We found that the regime of cortical networks in terms of their excitatory-inhibitory balance can modulate the formation and dynamics of assemblies. Networks with dominant excitatory interactions enabled a fast formation of assemblies, and this was accompanied by recruitment of other non-perturbed neurons, thus leading to some degree of nonspecific assembly formation. On the other hand, strong excitatory-inhibitory interaction recruited lateral inhibition, which slowed down the formation of assemblies but constrained them to the perturbed neurons. Our results suggest that these two regimes can be suitable for different computational and cognitive tasks with different trade-offs between speed and specificity. More generally, our work provides a framework to study network-wide behaviorally-relevant plasticity in biologically realistic networks.