2024/07/11 by Francesco Pinto, Nathalie Rauschmayr, Pinto, Francesco +7 · 1 citation
Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #Educational Technology and Assessment #FOS: Computer and information sciences #I.2.10 #I.2.7 #Intelligent Tutoring Systems and Adaptive Learning #K.4.1 #Machine Learning (cs.LG) #Online Learning and Analytics
paper · pdf · doi:10.48550/arxiv.2407.08707
openalex publication_date 2024/07/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Vision-Language Models (VLMs) have made remarkable progress in document-based Visual Question Answering (i.e., responding to queries about the contents of an input document provided as an image). In this work, we show these models can memorize responses for training samples and regurgitate them even when the relevant visual information has been removed. This includes Personal Identifiable Information (PII) repeated once in the training set, indicating these models could divulge memorised sensitive information and therefore pose a privacy risk. We quantitatively measure the extractability of information in controlled experiments and differentiate between cases where it arises from generalization capabilities or from memorization. We further investigate the factors that influence memorization across multiple state-of-the-art models and propose an effective heuristic countermeasure that empirically prevents the extractability of PII.