Synaptic Competition Enhances the Discrimination of Interfering Memories

Long-term plasticity is the primary mode of synaptic weight updating. However, other updating modes include modification of synaptic connections across neurons. The synaptic competition found during adult neurogenesis is an example. In this process, adult-born neurons are integrated into the existing neuronal pool by competing with mature neurons for synaptic connections from the entorhinal cortex. Evidence suggests that adult neurogenesis is critical for discriminating similar memories, i.e., memories with considerable interferences. This computational study aims to link synaptic competition with pattern separation. We show that synaptic competition and neuronal maturation are crucial for separating considerably overlapping memory patterns. Furthermore, we demonstrate that a feed-forward neural network trained by a competition-based learning rule can outperform a multi-layer perceptron trained by the backpropagation algorithm, especially when the sample size is small. These results unveil the functional implications and potential applications of synaptic competition in neural computation.