In real models with nonlinear activation functions, the update rules for opposing synapses are not mathematically identical, but research suggests perfect symmetry may not be essential; approximate symmetry emerging from independent learning is sufficient for effective network function.