Skip to content
AImpact
IT EN
Safety Intermediate Also known as: DP · Privacy differenziale

Differential privacy

A mathematical technique that adds controlled noise to training so that the presence or absence of a single individual in the dataset is not detectable from the model's output.

ShareLinkedInX

In practice

It is the de facto standard for models trained on health, tax, or messaging data. Apple, Google, and the US Census use it. It costs accuracy: more privacy means more noise.

Related terms

Data poisoning Fine-tuning

Seen in the wild

6 entries mentioning it
  1. Allen AI's Tülu 3: the first fully open post-training pipeline
    Medium
  2. UK AISI: the first government safety evaluations on GPT-4o and Claude 3.5
    High
  3. HuggingFace Accelerate 0.30: FSDP and DeepSpeed without extra code
    Medium
  4. Zephyr-7B: DPO on Mistral 7B beats Llama-2-70B-chat on MT-Bench
    High
  5. Hugging Face Inference Endpoints: deploy LLMs in two clicks
    Medium
  6. RAG: Retrieval-Augmented Generation enters the literature
    Landmark
← All terms