Game Theoretic Link Between STDP and Neural Computation

A postsynaptic neuron typically receives inputs from multiple presynaptic neurons whose connection strengths change according to spike-timing-dependent plasticity (STDP). Although many studies have examined the computational roles of specific circuit motifs or learning signals, a unified theoretical link between the STDP kernel and resulting circuit computation remains unclear. Here we introduce a game-theoretic framework offering a new perspective on a basic convergent circuit where one postsynaptic neuron integrates multiple inputs. We show that STDP dynamics are equivalent to a game hosted by the postsynaptic neuron, with each presynaptic neuron acting as an individual player. This game decomposes into three components—a potential game, a zero-sum game, and a non-strategic game—corresponding to contrastive PCA, lagged functional connectivity, and the initial connectivity pattern before learning. Our framework explains the evolution of such circuits as an interplay between collaboration for shared latent representation and competition in synaptic strength, providing new theoretical insight into neural computation in both biological and artificial networks.