Xiangjun (Ethan) Fu

Philadelphia, PA

xjf@seas.upenn.edu

Hi there! Welcome to my personal website!

I am an MSE student in Computer and Information Science at the University of Pennsylvania, broadly interested in understanding and improving the capabilities of Multimodal Large Language Models (MLLMs). In a field built on discrete autoregressive transformers, I explore how continuous generative modeling — normalizing flows, diffusion, VAEs — can unlock richer reasoning and planning, hopefully paving the road towards AGI. ;)

Currently, I work with Prof. Jiatao Gu on latent chain-of-thought reasoning with normalizing flows; our first paper, Latent Reasoning with Normalizing Flows, was accepted to the COLM 2026 Workshop on Efficient Reasoning — try the live demo! Before Penn, I earned my B.S. in Mathematics–Computer Science at UC San Diego, where I worked with PhD. Yupeng Hou and Prof. Julian McAuley on BLaIR, a benchmark for LLMs as semantic encoders in recommendation (ACL 2026, Oral).

Outside of research, you’ll find me on a soccer field ⚽, a pickleball court 🎾, or just about any sports venues!

news

Aug 08, 2026	Our paper “Latent Reasoning with Normalizing Flows” has been accepted to the COLM 2026 Workshop on Efficient Reasoning! Excited to meet yall at San Francisco in Oct! 🎉
Jul 16, 2026	Serving as a reviewer for EMNLP 2026 (ACL Rolling Review) and NeurIPS 2026! 📝
Jun 07, 2026	New preprint on arXiv: “Latent Reasoning with Normalizing Flows” — arXiv:2606.06447 📄
May 17, 2026	Our ACL 2026 paper “Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders” has been selected for an Oral presentation! 🎤
Apr 14, 2026	Our paper “Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders” has been accepted to ACL 2026 Main Conference! See yall at San Diego in July!

selected publications

COLM WS

Latent Reasoning with Normalizing Flows

Guancheng Tu^†, Xiangjun Fu^†, Suhao Yu, and 5 more authors

In COLM 2026 Workshop on Efficient Reasoning, 2026

arXiv Bib Demo

@inproceedings{tu2026nfcot,
  title = {Latent Reasoning with Normalizing Flows},
  author = {Tu, Guancheng and Fu, Xiangjun and Yu, Suhao and Tang, Yao and Kang, Haoqiang and Qin, Lianhui and Zhang, Yizhe and Gu, Jiatao},
  booktitle = {COLM 2026 Workshop on Efficient Reasoning},
  year = {2026},
}

ACL

Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders

Yupeng Hou^†, Jiacheng Li^†, Xiangjun Fu^†, and 4 more authors

In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL), 2026

arXiv Bib Oral

Oral Presentation

@inproceedings{fu2026blair,
  title = {Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders},
  author = {Hou, Yupeng and Li, Jiacheng and Fu, Xiangjun and He, Zhankui and Yan, An and Chen, Xiusi and McAuley, Julian},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL)},
  year = {2026},
}

arXiv

Larger Hausdorff Dimension in Scanning Pattern Facilitates Mamba-Based Methods in Low-Light Image Enhancement

Xinhua Wang, Caibo Feng, Xiangjun Fu, and 1 more author

arXiv preprint arXiv:2510.26001, 2025

arXiv Bib

@article{wang2025hausdorff,
  title = {Larger Hausdorff Dimension in Scanning Pattern Facilitates Mamba-Based Methods in Low-Light Image Enhancement},
  author = {Wang, Xinhua and Feng, Caibo and Fu, Xiangjun and Liu, Chunxiao},
  journal = {arXiv preprint arXiv:2510.26001},
  year = {2025},
}

news

selected publications

visitor flow