bitzj2015/EE599-Final-Project

EE599-Final-Project

Project Title

Obfuscating Web User Search Queries via Generative Adversarial Privacy

Team Member:

Jiang Zhang, Zhongxuan Ruan, Mengwei Yang

Team Roles:

• Jiang Zhang: Implement the basic GAP architecture, including one generator and one adver-sary.
• Zhongxuan Ruan: Prepossess AOL dataset and convert word tokens in queries into featurevectors.
• Mengwei Yang: Investigate DNN models in GAP architecture.
• All team members will work on the final presentation, slides, and report. And they will contribute equally to this project.

Required Environment:

• Pytorch 1.4, Python 3.6+, Ray ...
• Installation command:
  Pytorch: pip install torch torchvision
  Ray: pip install ray
  Other: pip install numpy matplotlib tqdm

Notes:

• Please run git pull and fix your local conflicts, before you upload your commits using git push.
• If you want to keep something locally without being uploaded to github, add them into .gitignore.

References:

• Deep Reinforcement Learning for Dialogue Generation.
• Toward Controlled Generation of Text.
• DCGAN tutorial.
• Seq2seq transformer tutorial.
• Incognito: A Method for Obfuscating Web Data.
• Generative Adversarial Privacy.
• SeqGAN Pytorch Implementation.
• Privacy- and Utility-Preserving Textual Analysis via CalibratedMultivariate Perturbations.
• Auditing Data Provenance in Text-Generation Models.
• Ray for multiprocessing.