Inhibitory Plasticity Enhances Sequence Storage Capacity and Retrieval Robustness

The generation of motor behaviors and the performance of complex cognitive tasks rely on sequential activity in specific brain structures. The mechanisms of learning and retrieval of these temporal patterns of activity are still poorly understood. Emerging evidence has high-lighted the importance of inhibition to learning and memory. However, the specific functions of inhibitory plasticity in the learning and retrieval of sequential activity have been studied very little, apart from its role in maintaining excitation-inhibition (E-I) balance. Using simulations and dynamical mean-field theory of balanced E-I networks, we found that sequences can be stored and retrieved using plasticity in both E-to-I and I-to-E pathways, in the absence of recurrent excitatory plasticity. Networks with both E-to-I and I-to-E plasticity are shown to exhibit higher optimal capacity than models in which plasticity is restricted to recurrent excitation. We further show that inhibitory plasticity enhances robustness to external noise and initial cue perturbation. Thus, our work suggests new computational roles for inhibitory plasticity in improving capacity and robustness of sequence learning.