Zixuan(Ian) Lan

zixuanlan@uchicago.edu, lanzixuan521@gmail.com

Hello! I'm an independent researcher. I received my Master's degree from the University of Chicago, where I worked with Professor Joe Zhou from Stony Brook University (life-long collaborator and advisor). My research focuses on improving model efficiency through algorithmic innovations. I am currently working on Large Language Models and Vision-Language Models, with interests spanning information compression (including token compression), architectural design, model interpretability, and Triton/CUDA optimization. I am actively learning GPU programming and am most familiar with the Ampere architecture.

GitHub Scholar CV

Publications

Text or Pixels? It Takes Half: On the Token Efficiency of Visual Text Inputs in Multimodal LLMs

Yanhong Li^*, Zixuan Lan^*, Joe Zhou. · EMNLP 2025

^*Equal contribution

Large language models (LLMs) and their multimodal variants can now process visual inputs, including images of text. This raises an intriguing question: can we compress textual inputs by feeding them as images to reduce token usage while preserving performance? In this paper, we show that visual text representationsare a practical and surprisingly effective form of input compression for decoder LLMs. Weexploit the idea of rendering long text inputs as a single image and provide it directly to the model. This leads to dramatically reduced number of decoder tokens required, offering a new form of input compression.

arXiv GitHub

Reduce What You Use: Input-Aware Matrix-Multiplication Pruning for LLMs

Zixuan Lan^*, Yanhong Li, Joe Zhou. · ICLR Under Review

Brief summary of the paper...

arXiv GitHub

Substituting from the Input: Distilling Sequential Computation in Transformer Language Models

Zixuan Lan^*, Jiaming Yang,Yanhong Li, Karen Livescu, Joe Zhou. · ICLR Under Review

Brief summary of the paper...

arXiv GitHub

ChunkOut: Information-Sufficient Token Pruning for Efficient Prompt Compression

Zixuan Lan^*, Yanhong Li^*, Joe Zhou.

^*Equal contribution

Brief summary of the paper...

arXiv GitHub