Data Contamination Calibration for Black-box LLMs

The rapid advancements of Large Language Models (LLMs) are tightly associated with the expansion of the training data size. However, the unchecked ultra-large-scale training sets introduce a series of potential risks like data contamination, i.e. the benchmark data is used for training. In this work, we propose a holistic method named Polarized Augment Calibration (PAC) along with a brand-new dataset named StackMIA to help detect the contaminated data and diminish the contamination effect. PAC extends the popular MIA (Membership Inference Attack) -from the machine learning community -by forming a more global target for detecting training data to clarify invisible training data. As a pioneering work, PAC is very much plug-and-play that can be integrated with most (if not all) current white-and black-box (for the first time) LLMs. By extensive experiments, PAC outperforms existing methods by at least 4.5%, in data contamination detection on more than 4 dataset formats, with more than 10 base LLMs. Besides, our application in real-world scenarios highlights the prominent presence of contamination and related issues. 1