Pandora's White-Box: Precise Training Data Detection and Extraction in Large Language Models

• URL: http://arxiv.org/abs/2402.17012v4
• Date: Mon, 15 Jul 2024 02:37:09 GMT
• Title: Pandora's White-Box: Precise Training Data Detection and Extraction in Large Language Models
• Authors: Jeffrey G. Wang, Jason Wang, Marvin Li, Seth Neel,
• Abstract summary: We develop state-of-the-art privacy attacks against Large Language Models (LLMs) New membership inference attacks (MIAs) against pretrained LLMs perform hundreds of times better than baseline attacks. In fine-tuning, we find that a simple attack based on the ratio of the loss between the base and fine-tuned models is able to achieve near-perfect MIA performance.
• Score: 4.081098869497239
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In this paper we develop state-of-the-art privacy attacks against Large Language Models (LLMs), where an adversary with some access to the model tries to learn something about the underlying training data. Our headline results are new membership inference attacks (MIAs) against pretrained LLMs that perform hundreds of times better than baseline attacks, and a pipeline showing that over 50% (!) of the fine-tuning dataset can be extracted from a fine-tuned LLM in natural settings. We consider varying degrees of access to the underlying model, pretraining and fine-tuning data, and both MIAs and training data extraction. For pretraining data, we propose two new MIAs: a supervised neural network classifier that predicts training data membership on the basis of (dimensionality-reduced) model gradients, as well as a variant of this attack that only requires logit access to the model by leveraging recent model-stealing work on LLMs. To our knowledge this is the first MIA that explicitly incorporates model-stealing information. Both attacks outperform existing black-box baselines, and our supervised attack closes the gap between MIA attack success against LLMs and the strongest known attacks for other machine learning models. In fine-tuning, we find that a simple attack based on the ratio of the loss between the base and fine-tuned models is able to achieve near-perfect MIA performance; we then leverage our MIA to extract a large fraction of the fine-tuning dataset from fine-tuned Pythia and Llama models. Our code is available at github.com/safr-ai-lab/pandora-llm.

Related papers

• Order of Magnitude Speedups for LLM Membership Inference [5.124111136127848]
  Large Language Models (LLMs) have the promise to revolutionize computing broadly, but their complexity and extensive training data also expose privacy vulnerabilities. One of the simplest privacy risks associated with LLMs is their susceptibility to membership inference attacks (MIAs) We propose a low-cost MIA that leverages an ensemble of small quantile regression models to determine if a document belongs to the model's training set or not.
  arXiv  Detail & Related papers  (2024-09-22T16:18:14Z)
• Accelerating Large Language Model Pretraining via LFR Pedagogy: Learn, Focus, and Review [50.78587571704713]
  Large Language Model (LLM) pretraining traditionally relies on autoregressive language modeling on randomly sampled data blocks from web-scale datasets. We take inspiration from human learning techniques like spaced repetition to hypothesize that random data sampling for LLMs leads to high training cost and low quality models which tend to forget data. In order to effectively commit web-scale information to long-term memory, we propose the LFR (Learn, Focus, and Review) pedagogy.
  arXiv  Detail & Related papers  (2024-09-10T00:59:18Z)
• Alpaca against Vicuna: Using LLMs to Uncover Memorization of LLMs [61.04246774006429]
  We introduce a black-box prompt optimization method that uses an attacker LLM agent to uncover higher levels of memorization in a victim agent. We observe that our instruction-based prompts generate outputs with 23.7% higher overlap with training data compared to the baseline prefix-suffix measurements. Our findings show that instruction-tuned models can expose pre-training data as much as their base-models, if not more so, and using instructions proposed by other LLMs can open a new avenue of automated attacks.
  arXiv  Detail & Related papers  (2024-03-05T19:32:01Z)
• Do Membership Inference Attacks Work on Large Language Models? [141.2019867466968]
  Membership inference attacks (MIAs) attempt to predict whether a particular datapoint is a member of a target model's training data. We perform a large-scale evaluation of MIAs over a suite of language models trained on the Pile, ranging from 160M to 12B parameters. We find that MIAs barely outperform random guessing for most settings across varying LLM sizes and domains.
  arXiv  Detail & Related papers  (2024-02-12T17:52:05Z)
• Scalable Extraction of Training Data from (Production) Language Models [93.7746567808049]
  This paper studies extractable memorization: training data that an adversary can efficiently extract by querying a machine learning model without prior knowledge of the training dataset. We show an adversary can extract gigabytes of training data from open-source language models like Pythia or GPT-Neo, semi-open models like LLaMA or Falcon, and closed models like ChatGPT.
  arXiv  Detail & Related papers  (2023-11-28T18:47:03Z)
• Army of Thieves: Enhancing Black-Box Model Extraction via Ensemble based sample selection [10.513955887214497]
  In Model Stealing Attacks (MSA), a machine learning model is queried repeatedly to build a labelled dataset. In this work, we explore the usage of an ensemble of deep learning models as our thief model. We achieve a 21% higher adversarial sample transferability than previous work for models trained on the CIFAR-10 dataset.
  arXiv  Detail & Related papers  (2023-11-08T10:31:29Z)
• Knowledge-Enriched Distributional Model Inversion Attacks [49.43828150561947]
  Model inversion (MI) attacks are aimed at reconstructing training data from model parameters. We present a novel inversion-specific GAN that can better distill knowledge useful for performing attacks on private models from public data. Our experiments show that the combination of these techniques can significantly boost the success rate of the state-of-the-art MI attacks by 150%.
  arXiv  Detail & Related papers  (2020-10-08T16:20:48Z)
• How Does Data Augmentation Affect Privacy in Machine Learning? [94.52721115660626]
  We propose new MI attacks to utilize the information of augmented data. We establish the optimal membership inference when the model is trained with augmented data.
  arXiv  Detail & Related papers  (2020-07-21T02:21:10Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.