Adversarial Learning of Privacy-Preserving and Task-Oriented Representations

Authors: Taihong Xiao, Yi-Hsuan Tsai, Kihyuk Sohn, Manmohan Chandraker, Ming-Hsuan Yang12434-12441

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the proposed method on face attribute prediction, showing that our method allows protecting visual privacy with a small decrease in utility performance. In addition, we show the utilityprivacy trade-off with different choices of hyperparameter for negative perceptual distance loss at training, allowing service providers to determine the right level of privacy-protection with a certain utility performance. Moreover, we provide an extensive study with different selections of features, tasks, and the data to further analyze their influence on privacy protection.
Researcher Affiliation Collaboration Taihong Xiao,1 Yi-Hsuan Tsai,2 Kihyuk Sohn,2 Manmohan Chandraker,2,3 Ming-Hsuan Yang1 1University of California, Merced 2NEC Laboratories America 3University of California, San Diego
Pseudocode No The paper includes a conceptual diagram (Figure 2) but does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement about releasing source code or a link to a code repository.
Open Datasets Yes We use the widely-used Celeb A (Liu et al. 2015) and MSCeleb-1M (Guo et al. 2016) datasets for experiments.
Dataset Splits Yes In most experiments, we split the Celeb A dataset into three parts, X1 with 160k images, X2 with 40k images, and the test set T with the rest.
Hardware Specification No The paper does not specify any hardware details (e.g., specific GPU or CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using a 'Res Net-50 model' but does not provide specific version numbers for software libraries, frameworks, or environments (e.g., PyTorch version, TensorFlow version, Python version).
Experiment Setup Yes Table 2: Results with different λ2 in the training stage. Other hyperparameters are fixed: λ1 = 1, μ1 = 0, μ2 = 1. We use the Res Net-50 model (He et al. 2016) for feature representation and two fully connected layers for latent classifier f. The Dec uses up-sampling layers to decode features to pixel images.