Fig. 6: Quantitative evaluation of gradient inversion resistance across neural architectures and biological modifications.
From: Concept transfer of synaptic diversity from biological to artificial neural networks
Bar plots compare mean squared reconstruction error (MSE) between baseline (green) and biomod (red) (MLP, AlexNet, ResNet20, ResNet32) on CIFAR10 dataset. Higher MSE values indicate better privacy protection, with biological modifications increasing reconstruction error by up to 56.3% in AlexNet and maintaining substantial improvements across all architectures. Black whiskers denote standard deviations.
