Differentially Private Learning Needs Hidden State (Or Much Faster Convergence)
–Neural Information Processing Systems
The main-stream analysis of privacy loss in DP-SGD is based on composition theorems, which quantify the total privacy loss of the training process across all its iterations.
Neural Information Processing Systems
Oct-1-2025, 21:16:34 GMT
- Country:
- Asia > Singapore (0.04)
- North America > United States
- California (0.04)
- Genre:
- Research Report (0.50)
- Industry:
- Information Technology > Security & Privacy (1.00)
- Technology: