Christopher Leckie

Toward Universal and Transferable Jailbreak Attacks on Vision-Language Models

POMP: A Theoretical Approach to Mitigate Forgetting in Finetuning Multi-Modal Models

Density-Aware Translation of Spurious Correlations in Zero-Shot VLMs

Leveraging Dark Knowledge for Intrinsic Multimodal Out-of-Distribution Detection

Geometry-Guided Adversarial Prompt Detection via Curvature and Local Intrinsic Dimension

GeoDetect: Geometric Adversarial Detection for VLPs

Characterising Universal Jailbreak Features and Refusal Direction in LLMs

Fortifying Time Series: DTW-Certified Robust Anomaly Detection

Open-Set Graph Anomaly Detection via Normal Structure Regularisation

CURVALID: A Geometrically-guided Adversarial Prompt Detection

GAD-VLP: Geometric Adversarial Detection for Vision-Language Pre-Trained Models