Abstract:The increasing integration of Visual Language Models (VLMs) into AI systems necessitates robust model alignment, especially when handling multimodal content that combines text and images. Existing evaluation datasets heavily lean towards text-only prompts, leaving visual vulnerabilities under evaluated. To address this gap, we propose \textbf{Text2VLM}, a novel multi-stage pipeline that adapts text-only datasets into multimodal formats, specifically designed to evaluate the resilience of VLMs against typographic prompt injection attacks. The Text2VLM pipeline identifies harmful content in the original text and converts it into a typographic image, creating a multimodal prompt for VLMs. Also, our evaluation of open-source VLMs highlights their increased susceptibility to prompt injection when visual inputs are introduced, revealing critical weaknesses in the current models' alignment. This is in addition to a significant performance gap compared to closed-source frontier models. We validate Text2VLM through human evaluations, ensuring the alignment of extracted salient concepts; text summarization and output classification align with human expectations. Text2VLM provides a scalable tool for comprehensive safety assessment, contributing to the development of more robust safety mechanisms for VLMs. By enhancing the evaluation of multimodal vulnerabilities, Text2VLM plays a role in advancing the safe deployment of VLMs in diverse, real-world applications.
Abstract:A popular strategy for active learning is to specifically target a reduction in epistemic uncertainty, since aleatoric uncertainty is often considered as being intrinsic to the system of interest and therefore not reducible. Yet, distinguishing these two types of uncertainty remains challenging and there is no single strategy that consistently outperforms the others. We propose to use a particular combination of probability and possibility theories, with the aim of using the latter to specifically represent epistemic uncertainty, and we show how this combination leads to new active learning strategies that have desirable properties. In order to demonstrate the efficiency of these strategies in non-trivial settings, we introduce the notion of a possibilistic Gaussian process (GP) and consider GP-based multiclass and binary classification problems, for which the proposed methods display a strong performance for both simulated and real datasets.