Yusen Peng

I am Yusen Peng, an incoming CSE PhD student at The Ohio State University (OSU), starting in Fall 2026. I am fortunate to have been advised by Dr. Sachin Kumar since my senior year as an undergraduate student at OSU 🌰!

My research interest is broadly efficient multimodal LLMs - How can we make these multimodal LLMs data efficient (e.g., token pruning, merging) and parameter efficient (e.g., SVD, MoE)? I am currently working on dynamic image tokenization to make vision language models (e.g., LLaVA, QwenVL) more efficient at inference time.

I earned my B.S. in Computer Science and Engineering from The Ohio State University with Summa Cum Laude and Honor Research Distinction. I completed my undergrduate thesis under the supervision of Dr. Alper Yilmaz on the topic of transformer-based video understanding (CascadeFormer). Previously, I worked on a variety of AI/ML topics including time series analysis (pytskit, TSB-FCST), 3D vision (SVD^3), and mechanistic interpretability (CE-Bench). You can check out my CV here for details.