Home Knowledge Base DP-SGD (Differentially Private Stochastic Gradient Descent)

DP-SGD (Differentially Private Stochastic Gradient Descent) is the foundational algorithm for training machine learning models with formal differential privacy guarantees — modifying standard SGD by clipping per-example gradients to bound sensitivity and adding calibrated Gaussian noise, ensuring that the trained model's parameters provably reveal limited information about any individual training example, enabling privacy-preserving deep learning on sensitive datasets.

What Is DP-SGD?

Why DP-SGD Matters

How DP-SGD Works

StepStandard SGDDP-SGD Modification
1. Sample BatchRandom mini-batchPoisson sampling (each example independently with probability q)
2. Compute GradientsPer-batch gradientPer-example gradients computed individually
3. ClipNo clippingClip each gradient to maximum norm C
4. AggregateSum gradientsSum clipped gradients
5. Add NoiseNo noiseAdd Gaussian noise N(0, σ²C²I)
6. Updateθ ← θ − η·gθ ← θ − η·(clipped_sum + noise)/batch_size

Key Parameters

Privacy Accounting

Practical Considerations

DP-SGD is the cornerstone of privacy-preserving deep learning — providing the only known method for training neural networks with rigorous mathematical privacy guarantees, making it indispensable for any application where model training on sensitive personal data must comply with privacy regulations.

dp-sgd (differentially private sgd)dp-sgddifferentially private sgdprivacy

Explore 500+ Semiconductor & AI Topics

From EUV lithography to CUDA optimization — search the full knowledge base or chat with our AI assistant.