Pegah Khayatan
Pegah Khayatan is a research engineer at ISIR, Sorbonne University, working under the supervision of Matthieu Cord. She is known for her research on representation steering and interpretability in vision-language-action models and multimodal large language models. She is a lead author of DiMaS, a distribution-matching steering strategy for flow-matching VLAs, and has also published work on analyzing finetuning representation shift for multimodal LLM steering. She holds an engineering degree from École Polytechnique and has interned at Valeo and the Vita Lab at EPFL.
“DiMaS: a Distribution-Matching Steering strategy tailored to flow-matching VLAs, which transports between representation distributions rather than shifting along a fixed direction”
Source→“behavioral features are linearly decodable but not linearly steerable, which motivates the distribution-matching design of DiMaS”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.