List your company
ResearchFormerly WizWand

Machine Unlearning on CIFAR-100 (forget set)

18.511Avg Increase in Forget AccuracySharpMinMax
5.730449.0484712.366515.68453Jun 16, 2025
Updated 8mo ago

Evaluation results

MethodLinks
SharpMinMaxRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0618.511
SharpMinMaxRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0617.5
SharpMinMaxRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0616.144
SharpMinMaxRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0615.589
SharpMinMaxRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0615.067
NGRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0614.7
NG + ASAMRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0613.389
NGRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0613.256
SharpMinMaxRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0613.1
NG + ASAMRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0613.022
NGRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0612.789
SharpMinMax + ASAMRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0612.711
SharpMinMax + ASAMRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0612.667
NG + ASAMRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0611.533
SharpMinMax + ASAMRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0611.2
NGRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0611.167
NG + ASAMRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0611.033
RLRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0610.989
SharpMinMax + ASAMRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.0610.667
NGRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.0610.333
RLRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.069.622
RLRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.069.589
RL + ASAMRelearn learning rate=0.004, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.069.467
NG + ASAMRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.069.289
SharpMinMax + ASAMRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.068.8
NGRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.068.644
RLRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.068.556
NG + ASAMRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.068.533
RLRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.068.5
RL + ASAMRelearn learning rate=0.003, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.068.489
RL + ASAMRelearn learning rate=0.004, Attack optimizer (A)=SGD, Model architecture=ResNet502025.068.367
SharpMinMax + ASAMRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.068.333
RL + ASAMRelearn learning rate=0.003, Attack optimizer (A)=SGD, Model architecture=ResNet502025.067.444
RL + ASAMRelearn learning rate=0.002, Attack optimizer (A)=ASAM, Model architecture=ResNet502025.067.378
RLRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.067.122
RL + ASAMRelearn learning rate=0.002, Attack optimizer (A)=SGD, Model architecture=ResNet502025.066.222