Christopher A. Choquette-Choo

research scientist@OpenAI·美国·OpenReview

研究方向

deep learning · natural language processing · deep neural networks · differential privacy · privacy auditing · data memorization · adversarial machine learning

Christopher A. Choquette-Choo

Language Models May Verbatim Complete Text They Were Not Explicitly Trained On

Exploring the limits of strong membership inference attacks on large language models

Recite, Reconstruct, Recollect: Memorization in LMs as a Multifaceted Phenomenon

Scalable Extraction of Training Data from Aligned, Production Language Models

Privacy Auditing of Large Language Models

Near-Exact Privacy Amplification for Matrix Mechanisms

Exploring and Mitigating Adversarial Manipulation of Voting-Based Leaderboards

The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD

Efficient and Privacy-Preserving Soft Prompt Transfer for LLMs

Scaling Laws for Differentially Private Language Models

POST: A Framework for Privacy of Soft-prompt Transfer

Phantom: General Trigger Attacks on Retrieval Augmented Language Generation

Privacy Amplification for Matrix Mechanisms

Teach LLMs to Phish: Stealing Private Information from Language Models

Correlated Noise Provably Beats Independent Noise for Differentially Private Learning

User Inference Attacks on Large Language Models

The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD