Optimizing Canaries for Privacy Auditing with Metagradient Descent
cs.LG, cs.CR
Submitted: 2025-07-21
Updated: 2026-09-22
Journal ref: International Conference on Learning Representations (ICLR) 2026, pp. 92994-93013
Code: https://github.com/oogle-deepmind/jax_privacy
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Automatic Differentiation of Algorithms for Machine Learning
- Unlocking High-Accuracy Differentially Private Image Classification through Scale
- CINIC-10 is not ImageNet or CIFAR-10
- Optimizing ML Training with Metagradient Descent
- How Well Can Differential Privacy Be Audited in One Run?
- Enhancing One-run Privacy Auditing with Quantile Regression-Based Membership Inference
- DARTS: Differentiable Architecture Search
- Auditing $f$-Differential Privacy in One Run
- Debugging Differential Privacy: A Case Study for Privacy Auditing
- Wide Residual Networks
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks