A predictive plasticity rule entails the anticipation of multiple spike sequences

Intelligent behavior depends on the brain’s ability to anticipate future events. Experimental evidence shows
reliable temporal sequences in the neural activity, suggesting a functional role of sequences for the association
and subsequent anticipation of events in time. However, how neurons can predict temporal patterns and fire
ahead of sensory inputs remains largely unknown. We propose a plasticity rule based on predictive processing,
where neurons learn a low-rank model of the synaptic input dynamics in their membrane potential. Neurons
thereby amplify those synapses that maximally predict other synaptic inputs based on their temporal relations.
We show that neurons predict sequences over long timescales and shift their spikes toward the first inputs in a
sequence. Moreover, we show how the predictive plasticity rule entails the prediction of multiple sequences
in a recurrent neural network. Neurons in the network selectively anticipate sequences in the input spike
trains via a recurrent inhibition mechanism. In sum, our results indicate prediction as a guiding principle
for the anticipation of multiple spike sequences in both single neurons and recurrent neural networks. Our
model shows that the existence of reliable spike sequences can be explained by predictive mechanisms at the
single-neuron level.