SPQR: A Multi-Dimensional Benchmark for Safety Alignment under Benign Model Adaptation
A benchmark for testing whether text-to-image safety alignment still holds after benign fine-tuning and deployment-time adaptation.
Jun 1, 2026
Improving LLM First-Token Predictions in Multiple-Choice Question Answering via Output Prefilling
A simple prefilling strategy that makes first-token probability evaluation more reliable for multiple-choice LLM benchmarks.
Jan 1, 2026
Unlearning Vision Transformers without Retaining Data via Low-Rank Decompositions
A low-rank unlearning approach for Vision Transformers that avoids retaining the original training data.
Dec 1, 2024
Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks
Fine-tuning attacks in one language can break multilingual safety alignment, suggesting safety information is partly language-agnostic.
Oct 1, 2024
Safe-CLIP: Removing NSFW Concepts from Vision-and-Language Models
A method for reducing unsafe concept associations in CLIP while preserving useful vision-language behavior.
Nov 27, 2023
Unveiling the Impact of Image Transformations on Deepfake Detection: An Experimental Analysis
An experimental study of how common image transformations affect the robustness of deepfake detectors.
Sep 5, 2023
Towards Explainable Navigation and Recounting
An explainable visual-navigation setting where an agent recounts indoor scenes and links generated language to visual explanation maps.
Sep 5, 2023
Revisiting The Evaluation of Class Activation Mapping for Explainability: A Novel Metric and Experimental Analysis
A reproducibility-focused evaluation metric for Class Activation Mapping methods in computer vision.
Jun 1, 2021