关键词: bump attractor, synaptic plasticity, robustness, sensitivity

Abstract: The way in which persistent firing activity and synaptic plasticity are orchestrated to underlie working memory in recurrent neural networks is not fully understood. Here, we build a continuous attractor network of pyramidal cells and interneurons to simulate an oculomotor delayed response task. Both short-term facilitation (STF) and short-term depression (STD) manifest at synapses between pyramidal cells. The efficacy of individual synapses depends on the time constants of STF and STD as well as the presynaptic firing rate. Self-sustained firing activity (i.e., a bump attractor) during the delay period encodes the cue position. The bump attractor becomes more robust against random drifts and distractions with enhancing STF or reducing STD. Keeping STF and STD at appropriate levels is crucial for optimizing network performance. Our results suggest that, besides slow recurrent excitation and strong global inhibition, short-term plasticity plays a prominent role in facilitating mnemonic behavior.

Key words: bump attractor, synaptic plasticity, robustness, sensitivity