I am Yusen Peng, a final year undergraduate Computer Science and Engineering (CSE) student at The Ohio State University (OSU), graduating in Spring 2026.

I am actively seeking PhD position in Fall 2026 in vision-language models and multimodal learning.

Publications 🚀

CE-Bench: Towards a Reliable Contrastive Evaluation Benchmark of Interpretability of Sparse Autoencoders

Alex Gulko*, Yusen Peng*, Sachin Kumar.
Proceedings of the 8th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
[Arxiv] | [ACL Anthology] | [Github] | [HuggingFace]

CascadeFormer: A Family of Two-stage Cascading Transformers for Skeleton-based Human Action Recognition

Yusen Peng, Alper Yilmaz.
arXiv preprint 2025; under review
[Arxiv] | [Github] | [HuggingFace]

Lower-quality public housing corresponds to elevated flood risk and social disadvantage

Woi Sok Oh, Kelsea Best, Meri Davlasheridze, Yusen Peng
Under review at Earth’s Future
[Paper]

SIGNAL: A Comprehensive Time Series Library

Advisor: Dr. John Paparrizos
Under preparation for JMLR

Ongoing Projects ⭐️

Dynamic Patch Pooling for Efficient Vision Transformers

Advisor: Dr. Sachin Kumar
[Github]

CascadeFormer 2: Towards a Cost-Aligned Anomaly Detection Agent via Reinforcement Learning with Verifiable Rewards

Advisor: Dr. Alper Yilmaz
[Github]

SVD-π3: Efficient Visual Geometry Learning via Singular Value Decomposition

Mentor: Haoxuan Wang; Advisor: Dr. Yan Yan
[Github]

Awards 🏅

• Honorable Mention, CRA Outstanding Undergraduate Researcher Award (2025)
  • by Computing Research Association (CRA)
• Undergraduate Research Scholarship 2025 @ OSU
  • $2,200 scholarship awarded based on my Honors Thesis proposal @ OSU

Academic Service 📚

• Conference Reviewer @ NeurIPS
  • reviewed 3 papers at NeurIPS 2025 Mechanistic Interpretability Workshop

Campus Poster Presentations 👨‍👨‍👦‍👦

• SIGNAL: A Comprehensive Time Series Analysis Library
  • The DATUM Lab | Advisor: Dr. John Paparrizos
  • Poster presented at the CSE Annual Research Expo 2025 at The Ohio State University
• A Comparison of CSV, HDF5, Zarr, and netCDF4 in Performing Common I/O Operations
  • Advisor: Dr. Suren Byna; GitHub
  • Poster presented at the CSE Annual Research Expo 2024 at The Ohio State University

AI/ML Competitions 💡

• NASA Airport Throughput Prediction Challenge 2024 (Top 10 Finalist)
  • Ranked 9th out of 51 teams in the final Leaderboard