Major efforts in neuroscience are dedicated to uncovering synaptic plasticity rules which underlie learning-related changes of neural activity in vivo. Plasticity rules measured in vitro are highly heterogeneous, suggesting that learned behaviors could be supported by particular combinations of plasticity rules. We studied the emergence of preparatory neural activity in the anterior lateral motor (ALM) cortex during a delayed-discrimination task, a paradigmatic example of learned neural-dynamics supporting flexible goal-directed behaviors. Specifically, we constrained an attractor network model to capture the empirical delay-period dynamics and time-dependent susceptibility to distracting stimuli. Our analytical mean-field-theory solutions of the attractor dynamics exhibited by the network reveal that a particular combination of symmetric and temporally-asymmetric plasticity rules is necessary to capture the observed neural dynamics. We extracted the network's effective feed-forward connectivity structure, and studied its dependence on the characteristics of the plasticity rules. This analysis identified direct links between the plasticity rules in-effect and the dynamical features of delay-period neural activity. Used together with recent cellular-resolution measurements of functional connectivity in vivo, our modeling allows us to relate specific changes of network connectivity to the learned behaviors they support. Learned neural activity that emerges based on distinct plasticity mechanisms may facilitate flexible and generalizable assignment of behavioral responses to sensory stimuli. For example, our work may help understand how circuits can, with minimal reorganization, generate the same motor plan in response to different sensory stimuli.