/Neurips_2024_code

Adversarial Vulnerability as a Consequence of Manifold Inseparibility

Primary LanguagePython

Command for ADAM clean training: python -u convergence.py --data ${D} --seed ${SEED} --epochs 1000

Command for clean training with batch normalization: python -u convergence_bn.py --data ${D} --seed ${SEED} --epochs 1000

Command for clean training with second-order optimization using KFAC: python -u mnist_kfac.py --data ${D} --seed ${SEED} --epochs 1000

Arguments: ${D}="MNIST" or "FashionMNIST"