Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Min-K%++: Improved Baseline for Detecting Pre-Training Data from Large Language Models

About

The problem of pre-training data detection for large language models (LLMs) has received growing attention due to its implications in critical issues like copyright violation and test data contamination. Despite improved performance, existing methods (including the state-of-the-art, Min-K%) are mostly developed upon simple heuristics and lack solid, reasonable foundations. In this work, we propose a novel and theoretically motivated methodology for pre-training data detection, named Min-K%++. Specifically, we present a key insight that training samples tend to be local maxima of the modeled distribution along each input dimension through maximum likelihood training, which in turn allow us to insightfully translate the problem into identification of local maxima. Then, we design our method accordingly that works under the discrete distribution modeled by LLMs, whose core idea is to determine whether the input forms a mode or has relatively high probability under the conditional categorical distribution. Empirically, the proposed method achieves new SOTA performance across multiple settings. On the WikiMIA benchmark, Min-K%++ outperforms the runner-up by 6.2% to 10.5% in detection AUROC averaged over five models. On the more challenging MIMIR benchmark, it consistently improves upon reference-free methods while performing on par with reference-based method that requires an extra reference model.

Jingyang Zhang, Jingwei Sun, Eric Yeats, Yang Ouyang, Martin Kuo, Jianyi Zhang, Hao Frank Yang, Hai Li• 2024

Related benchmarks

TaskDatasetResultRank
Membership Inference AttackWikiMIA length 64
AUC0.8671
84
Membership InferenceWikiMIA 32 tokens 1.0
ROC AUC84.8
66
Suffix RankingExtraction Challenge Dataset
MP (%)47.6
66
Membership Inference AttackPile-CC
TPR @ 1%1.3
61
Membership Inference AttackAsclepius (fine-tuned)
TPR@FPR=0.011.82
58
Membership Inference AttackWikiMIA length 128
AUC84.33
56
Membership Inference AttackWikiMIA length 32
TPR @ 5% FPR0.3501
54
Membership Inference AttackWikipedia
AUC0.642
52
Membership Inference AttackMIMIR Github
AUC73.8
52
Membership Inference AttackWikiMIA length 128 (test)
TPR @ 5% FPR42.62
49
Showing 10 of 77 rows
...

Other info

Follow for update