Lei Li

Carnegie Mellon University

I am an Assistant Professor in the Language Technologies Institute of the School of Computer Science at Carnegie Mellon University. I am also affiliated with CMU CyLab and University of California Santa Barbara Computer Science Department. I work on generative AI for language and science, including multilingual NLP, machine translation (text, speech), security of large language models, agentic LLM, and AI for drug discovery and protein design.

  • LLM Reasoning (weak LLM helps strong LLM). [arxiv]
  • LLM Watermark: Unigram-Watermark to detect AI generated text, and GINSEW to defend against model extraction attack.
  • CGMH: A method for controllable text generation from specified keywords. [arxiv]
  • Analyzing anisotropic sentence embeddings from pre-trained language models. [arxiv]
  • Data-efficient methods for many-to-many neural machine translation. [mRASP, mRASP2]
  • Glancing Transformer (GLAT): non-autoregressive translation models are equally good as autoregressive Transformer. [arxiv]
  • VOLT: Learning vocabulary via optimal transport (ACL 2021 Best Paper). [arxiv]
  • SOLO: Fast and accurate object instance segmentation. [SOLO, SOLOv2]