Nathan L Clement

Automatic Hard Example Synthesis with Multi-Level Agentic Data Curation

Genglin Liu, Muye Zhang, Krishnamurthy Viswanathan, Nichole J. Hansen, Blaž Bratanič, Nathan L Clement, et al.

Multimodal Large Language Models (MLLMs) are increasingly deployed for nuanced content safety and moderation tasks, yet they remain vulnerable to adversarial attacks and out-of-distribution edge cases. Traditional active learning and manual annotation fail to scale against the co…Read more