ConfounderGAN: Protecting Image Data Privacy with Causal Confounder

Authors: Qi Tian, Kun Kuang, Kelu Jiang, Furui Liu, Zhihua Wang, Fei Wu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments are conducted in six image classification datasets, consisting of three natural object datasets and three medical datasets.
Researcher Affiliation Collaboration 1 College of Computer Science and Technology, Zhejiang University, Hangzhou, China 2 Huawei Noah s Ark Lab, Beijing, China 3 Shanghai Institute for Advanced Study of Zhejiang University, Shanghai, China 4 Shanghai AI Laboratory, Shanghai, China 5 Key Laboratory for Corneal Diseases Research of Zhejiang Province, Hangzhou, China
Pseudocode Yes Algorithm 1: Minibatch training of Confounder GAN
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We provide instructions needed to reproduce our experimental results in Appendix G.
Open Datasets Yes We select three natural object datasets and three medical datasets for algorithm evaluation, including SVHN [22], CIFAR10 [16], Image Net25 [5], Blood MNIST [45], Keratitis and ISIC [1].
Dataset Splits Yes Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] See Appendix G.
Hardware Specification Yes Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Appendix G.
Software Dependencies No No specific version numbers for software dependencies were mentioned in the main text. Appendix G is referenced for more experimental details, but the self-assessment only confirms training details, not specific software versions.
Experiment Setup Yes Referring to Huang et al. [12], the maximum perturbation is set to 8/255 in SVHN and CIFAR10, and 16/255 in other datasets, thus ensuring it is imperceptible to human observers. These settings are fixed for all experiments, unless otherwise explicitly stated. See Appendix G for more experimental details.