Replica-mean-field limits for intensity-based neural networks