Private PoEtry: Private In-Context Learning via Product of Experts

TL;DR

This paper introduces a theoretically grounded, parallelizable method for private in-context learning using a Product-of-Experts model, significantly improving accuracy while preserving privacy across multiple tasks.

Contribution

The paper proposes a novel Product-of-Experts framework for private in-context learning, addressing limitations of existing methods with a more effective and parallelizable approach.

Findings

Improves accuracy by over 30 percentage points on average.

Maintains strong privacy guarantees across diverse datasets.

Efficiently parallelizable algorithm.

Abstract

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.