STARE uses step-wise RL to attack multimodal models, achieving 68% higher attack success rate while revealing that adversarial optimization concentrates conceptual toxicity early and detail toxicity late in the generation trajectory.
The power of scale for parameter-efficient prompt tuning
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
verdicts
UNVERDICTED 4roles
background 1polarities
background 1representative citing papers
CAAT selects critical parameters for adversarial robustness in ViTs and applies PEFT to tune only those, yielding a 4.3% robustness drop versus full AT while using ~6% of parameters.
LLMs share task-specific attention heads across prompting styles, with activation strength explaining performance differences and failures arising from competing representations.
PaliGemma is an open 3B VLM based on SigLIP and Gemma that achieves strong performance on nearly 40 diverse open-world tasks including benchmarks, remote-sensing, and segmentation.
citing papers explorer
-
STARE: Step-wise Temporal Alignment and Red-teaming Engine for Multi-modal Toxicity Attack
STARE uses step-wise RL to attack multimodal models, achieving 68% higher attack success rate while revealing that adversarial optimization concentrates conceptual toxicity early and detail toxicity late in the generation trajectory.
-
Efficient Adversarial Training via Criticality-Aware Fine-Tuning
CAAT selects critical parameters for adversarial robustness in ViTs and applies PEFT to tune only those, yielding a 4.3% robustness drop versus full AT while using ~6% of parameters.
-
Shared Lexical Task Representations Explain Behavioral Variability In LLMs
LLMs share task-specific attention heads across prompting styles, with activation strength explaining performance differences and failures arising from competing representations.
-
PaliGemma: A versatile 3B VLM for transfer
PaliGemma is an open 3B VLM based on SigLIP and Gemma that achieves strong performance on nearly 40 diverse open-world tasks including benchmarks, remote-sensing, and segmentation.