Differentially Private Adaptive Optimization with Delayed Preconditioners
This addresses the challenge of efficient private training for machine learning practitioners, offering a solution without needing extra data, though it is incremental by building on prior adaptive optimization techniques.
The paper tackles the problem of privacy noise undermining adaptive optimizers in differentially private model training by proposing DP^2, a method that uses delayed preconditioners to reduce noise and improve convergence, achieving up to 4x faster convergence and matching state-of-the-art methods that require auxiliary data.
Privacy noise may negate the benefits of using adaptive optimizers in differentially private model training. Prior works typically address this issue by using auxiliary information (e.g., public data) to boost the effectiveness of adaptive optimization. In this work, we explore techniques to estimate and efficiently adapt to gradient geometry in private adaptive optimization without auxiliary data. Motivated by the observation that adaptive methods can tolerate stale preconditioners, we propose differentially private adaptive training with delayed preconditioners (DP^2), a simple method that constructs delayed but less noisy preconditioners to better realize the benefits of adaptivity. Theoretically, we provide convergence guarantees for our method for both convex and non-convex problems, and analyze trade-offs between delay and privacy noise reduction. Empirically, we explore DP^2 across several real-world datasets, demonstrating that it can improve convergence speed by as much as 4x relative to non-adaptive baselines and match the performance of state-of-the-art optimization methods that require auxiliary data.