Corpus ID: 167217888

Overlearning Reveals Sensitive Attributes

@article{Song2020OverlearningRS,
  title={Overlearning Reveals Sensitive Attributes},
  author={Congzheng Song and Vitaly Shmatikov},
  journal={ArXiv},
  year={2020},
  volume={abs/1905.11742}
}
"Overlearning" means that a model trained for a seemingly simple objective implicitly learns to recognize attributes and concepts that are (1) not part of the learning objective, and (2) sensitive from a privacy or bias perspective. For example, a binary gender classifier of facial images also learns to recognize races\textemdash even races that are not represented in the training data\textemdash and identities. We demonstrate overlearning in several vision and NLP models and analyze its… Expand
24 Citations
FaceLeaks: Inference Attacks against Transfer Learning Models via Black-box Queries
  • 1
  • PDF
Does Learning Stable Features Provide Privacy Benefits for Machine Learning Models?
  • Highly Influenced
  • PDF
Quantifying and Mitigating Privacy Risks of Contrastive Learning
  • Highly Influenced
  • PDF
Attribute Privacy: Framework and Mechanisms
  • PDF
High-Fidelity Extraction of Neural Network Models
  • 25
Information Leakage in Embedding Models
  • 11
  • PDF
Obfuscation of Images via Differential Privacy: From Facial Images to General Images
  • PDF
...
1
2
3
...

References

SHOWING 1-10 OF 47 REFERENCES
Censoring Representations with an Adversary
  • 257
  • PDF
Adversarial Removal of Demographic Attributes from Text Data
  • 104
  • PDF
Towards Robust and Privacy-preserving Text Representations
  • 51
  • PDF
Minimax Filter: Learning to Preserve Privacy from Inference Attacks
  • Jihun Hamm
  • Computer Science, Mathematics
  • J. Mach. Learn. Res.
  • 2017
  • 48
  • PDF
Privacy Partitioning: Protecting User Data During the Deep Learning Inference Phase
  • 9
  • PDF
Exploiting Unintended Feature Leakage in Collaborative Learning
  • 278
  • PDF
Privacy-preserving Neural Representations of Text
  • 29
  • PDF
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
  • 400
  • PDF
Age Progression/Regression by Conditional Adversarial Autoencoder
  • Zhifei Zhang, Yang Song, H. Qi
  • Computer Science, Mathematics
  • 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2017
  • 389
  • PDF
Invariant Representations without Adversarial Training
  • 66
  • Highly Influential
  • PDF
...
1
2
3
4
5
...