A new computational study shows that a constrained Hebbian learning rule achieves lower representational cost than backpropagation and target propagation under biologically motivated sparsity ...