Vitor R. Carvalho | Searcharxiv

arXiv2026

VLMGuard: Bootstrapping Malicious Prompt Detectors from Unlabeled Vision-Language Prompts in the Wild

Vision-language Models (VLMs) are essential for contextual understanding of both visual and textual information. However, their vulnerability to adversarially manipulated inputs presents significant risks, leading to compromised outputs and raising concerns about the reliability in VLM-integrated applications. Detecting these malicious prompts is thus crucial for maintaining trust in VLM generations. A major challenge in developing a safeguarding prompt classifier is the lack of a large amount of labeled benign and malicious data. To address the issue, we introduce VLMGuard, a novel learning framework that leverages the unlabeled user prompts in the wild for malicious prompt detection. These unlabeled prompts, which naturally arise when VLMs are deployed in the open world, consist of both benign and malicious information. To harness the unlabeled data, we present an automated maliciousness estimation score for distinguishing between benign and malicious samples within this unlabeled mixture, thereby enabling the training of a binary prompt classifier on top. Notably, our framework does not require extra human annotations and is robust to realistic prompt variations, offering strong flexibility and practicality for real-world applications. Extensive experiments show that VLMGuard achieves superior detection results, improving AUROC by 5.39% on average over the state-of-the-art method. Disclaimer: This paper may contain offensive examples; reader discretion is advised. Code is available at: https://github.com/radiolab-ntu/vlmguard.

cs.LG↗

arXiv2020

Improving Robustness on Seasonality-Heavy Multivariate Time Series Anomaly Detection

Robust Anomaly Detection (AD) on time series data is a key component for monitoring many complex modern systems. These systems typically generate high-dimensional time series that can be highly noisy, seasonal, and inter-correlated. This paper explores some of the challenges in such data, and proposes a new approach that makes inroads towards increased robustness on seasonal and contaminated data, while providing a better root cause identification of anomalies. In particular, we propose the use of Robust Seasonal Multivariate Generative Adversarial Network (RSM-GAN) that extends recent advancements in GAN with the adoption of convolutional-LSTM layers and attention mechanisms to produce excellent performance on various settings. We conduct extensive experiments in which not only do this model displays more robust behavior on complex seasonality patterns, but also shows increased resistance to training data contamination. We compare it with existing classical and deep-learning AD models, and show that this architecture is associated with the lowest false positive rate and improves precision by 30% and 16% in real-world and synthetic data, respectively.

cs.LG↗