Table 2.

Unique feature memorisation occurs frequently in neural networks.

Dataset	Model	Number of runs	Proportion of memorised networks	Average M score
Two moons	MLP-2	1000	65%	0.51
F-MNIST	CNN-1	10	20%	0.4
CIFAR-10	ResNet18	100	46%	0.03
Celeb-A	ResNet18	10	80%	0.01
CheXpert	DenseNet121	10	60%	0.007

Here we show average M scores for a range of datasets and model architectures.