PhD student, University of Roma "La Sapienza"
2 papers at NeurIPS 2025
Attention heads in text-generative models specialize in semantic and visual concepts. Leveraging this property, we can reliably suppress or enhance specific attributes in both language and vision-language tasks.
We propose an inference-time intervention framework based on Optimal Transport that generalizes previous methods and allows interpretable control of both LLMs and Diffusion models.
