interpretabilityfactneutralHebbian learning achieves lower representational cost than Dense Difference Target Propagation and backpropagation under matched sparsity constraintsNeural and Evolutionary Computing28 Jul 2026http://arxiv.org/abs/2607.16027v1