Stealing various attributes of a blackbox ML model has been recently gaining popularity: parameters [35], hyperparameters [37], architecture [24], information on training data [31] and decision boundaries [25].
Zip files (containing pytorch checkpoint, transferset pickle file, hyperparameters and logs) can be downloaded using the links below. Specifically, the knockoffs are resnet34s at B=60k using imagenet as the query set ($P_A$). $F_V$RandomAdaptive Caltech256 zip (76.0%) zip (%) CUBS...
Zip files (containing resnet-34 pytorch checkpoint.pth.tar, hyperparameters and training logs): Caltech256(Accuracy = 78.4%) CUBS200(77.1%) Indoor67(76.0%) Diabetic5(59.4%) #Format:$ python knockoff/victim/train.py DS_NAME ARCH -d DEV_ID \ -o models/victim/VIC_DIR -e EPOCHS --pret...