Gap-K%: Measuring Top-1 Prediction Gap for Detecting Pretraining Data (2026.acl-long)
Copied to clipboard
| Challenge: | Existing state-of-the-art methods for pretraining data are largely undisclosed, resulting in ethical and copyright concerns. |
| Approach: | They propose a method that leverages the log probability gap between the top-1 predicted token and the target token, incorporating a sliding window strategy to capture local correlations and mitigate token-level fluctuations. |
| Outcome: | The proposed method outperforms baselines on WikiMIA and MIMIR benchmarks and achieves state-of-the-art performance. |
Similar Papers
Pretraining Data Detection for Large Language Models: A Divergence-based Calibration Method (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to detect text in training corpus are limited due to their low token probabilities. |
| Approach: | They propose a method to calibrate token probabilities for pretraining data detection by using a divergence-based calibration method. |
| Outcome: | The proposed method significantly outperforms existing methods on Chinese text on English-language benchmarks and patents. |
Con-ReCall: Detecting Pre-training Data in LLMs via Contrastive Decoding (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods analyze training data with member and non-member contexts, overlooking potential insights from both member and not-member. |
| Approach: | They propose a method that leverages asymmetric distributional shifts induced by member and non-member contexts through contrastive decoding to enhance membership inference. |
| Outcome: | The proposed approach outperforms the current state-of-the-art on the WikiMIA benchmark and is robust against various text manipulation techniques. |
Tag&Tab: Pretraining Data Detection in Large Language Models Using Keyword-Based Membership Inference Attack (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies on detecting pretraining data in large language models have focused on sentence-level membership inference attacks (MIAs) but these methods often exhibit poor accuracy, failing to account for the semantic importance of textual content and word significance. |
| Approach: | They propose a method that leverages established natural language processing techniques to tag keywords in input text and then uses them to obtain probabilities and calculate their average log-likelihood to determine input text membership. |
| Outcome: | The proposed method exploits established natural language processing techniques to tag keywords in input text and calculate their average log-likelihood to determine input text membership. |
Improving Pretraining Techniques for Code-Switched NLP (2023.acl-long)
Copied to clipboard
| Challenge: | Multilingual pretraining models for code-switched inputs are a key component of NLP applications. |
| Approach: | They propose to use masked language modeling techniques to mask code-switched text that are cognizant of language boundaries prior to masking. |
| Outcome: | The proposed techniques improve performance on two downstream tasks, Question Answering (QA) and Sentiment Analysis (SA), compared to standard pretraining techniques. |
Gradient Ascent Post-training Enhances Language Model Generalization (2023.acl-short)
Copied to clipboard
| Challenge: | Recent studies show that language models can perform diverse downstream NLP tasks in a zero-shot manner without any gradient updates. |
| Approach: | They show that updating pretrained LMs with Gradient Ascent Post-training enhances zero-shot generalization capabilities without any task-specific fine-tuning. |
| Outcome: | The proposed method can be comparable to 2-3x times larger LMs across 12 different NLP tasks. |
Identifying Pre-training Data in LLMs: A Neuron Activation-Based Detection Framework (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for detecting pre-training data in large language models rely on superficial features like prediction confidence and loss, resulting in mediocre performance. |
| Approach: | They propose a new algorithm to analyze neuron activation patterns between training and non-training data in large language models to improve their performance. |
| Outcome: | The proposed algorithm outperforms existing methods across three benchmarks and multiple LLMs. |
Context-Aware Membership Inference Attacks against Pre-trained Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Prior Membership Inference Attacks on pre-trained Large Language Models fail at LLMs due to ignoring the generative nature of LLM data. |
| Approach: | They propose a method that adapts MIA statistical tests to the perplexity dynamics of subsequences within a data point. |
| Outcome: | The proposed method significantly outperforms prior approaches, revealing context-dependent memorization patterns in pre-trained LLMs. |
Probing Critical Learning Dynamics of PLMs for Hate Speech Detection (2024.findings-eacl)
Copied to clipboard
| Challenge: | Existing studies on pretrained language models (PLMs) for hate speech detection have not investigated how their performance is affected by pretraining and finetuning. |
| Approach: | They propose to compare pretrained language models, evaluate their seed robustness, finetuning settings, and the impact of pretraining data collection time. |
| Outcome: | The proposed models show that they are more robust than other models and that they have a better chance of performing better than domain-specific models. |
Establishing a Scale for Kullback-Leibler Divergence in Language Models Across Various Settings (2026.findings-acl)
Copied to clipboard
| Challenge: | Fig. 1 and 2 shows that log-likelihood vectors provide a consistent representation for language models . weight permutation symmetries and architectural dependencies hinder direct comparisons between models with different learning methods or designs. |
| Approach: | They propose a log-likelihood vector for comparing language models as probability distributions . they establish a consistent scale for KL divergence across various settings . |
| Outcome: | The proposed model comparisons show that the log-likelihood space is smaller than the weight space . the proposed model compares language models across checkpoints, model sizes, quantization, fine-tuning, and layers . |
Forget What You Know about LLMs Evaluations - LLMs are Like a Chameleon (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) excel on public benchmarks, but high scores may mask overreliance on dataset-specific surface cues rather than true language understanding. |
| Approach: | They propose a meta-evaluation framework that systematically rephrases benchmark inputs to detect overfitting. |
| Outcome: | The proposed framework detects performance degradation indicative of superficial pattern reliance on dataset-specific cues and distortion levels. |