Private PoEtry: Private In-Context Learning via Product of Experts
This addresses privacy concerns for users of in-context learning in sensitive applications like text classification, math, and vision-language tasks, representing a novel method for a known bottleneck.
The paper tackles the problem of privacy leakage in in-context learning for large language models by proposing a Product-of-Experts framework, resulting in over 30 percentage points average accuracy improvement compared to prior methods while maintaining strong privacy guarantees.
In-context learning (ICL) enables Large Language Models (LLMs) to adapt to new tasks with only a small set of examples at inference time, thereby avoiding task-specific fine-tuning. However, in-context examples may contain privacy-sensitive information that should not be revealed through model outputs. Existing differential privacy (DP) approaches to ICL are either computationally expensive or rely on heuristics with limited effectiveness, including context oversampling, synthetic data generation, or unnecessary thresholding. We reformulate private ICL through the lens of a Product-of-Experts model. This gives a theoretically grounded framework, and the algorithm can be trivially parallelized. We evaluate our method across five datasets in text classification, math, and vision-language. We find that our method improves accuracy by more than 30 percentage points on average compared to prior DP-ICL methods, while maintaining strong privacy guarantees.