Research scientist, Apple
1 paper at NeurIPS 2025
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.