Differentially Private Steering for Large Language Model Alignment

2025-01-22 · ICLR 2025 Poster · anchor

Findings

IC-459Qwen-2.5 models show that the privacy-utility tradeoff for differentially private steering improves with model size
IC-460Non-private activation steering of Llama-2-7B and Qwen-2.5-7B leaks membership information from the alignment dataset, while PSA reduces the empirical privacy loss
IC-461Adding calibrated Gaussian noise to steering vectors (PSA) preserves alignment performance comparable to non-private mean steering across Llama-2-7B, Mistral-7B, Gemma-2-2B, and Qwen-2.5-7B