Privacy of the last iterate in cyclically-sampled DP-SGD on nonconvex composite losses.
Weiwei KongMónica RiberoPublished in: CoRR (2024)
Keyphrases
- dynamic programming
- privacy preserving
- objective function
- convex optimization
- personal information
- private information
- privacy aware
- statistical databases
- personal data
- stochastic gradient descent
- sensitive information
- security issues
- privacy preserving data mining
- private data
- quadratic optimization problems
- legal issues
- privacy issues
- global optimization
- cost function
- differential privacy
- information loss
- data publishing
- location privacy