CV

Graduate Researcher

GMLR Lab, University of Pennsylvania

Advised by Prof. Jiatao Gu.

• Designing a continuous latent chain-of-thought reasoning pipeline on a Qwen3-8B backbone that operates in a lower-dimensional space while preserving semantic structure.
• Fine-tuning an LLM-based VAE encoder to provide semantic guidance for latent thinking-token generation; using normalizing flows to autoregressively generate thinking tokens under an MSE objective against VAE-encoded latents.
• Computing step-wise log-likelihoods to enable reinforcement learning over reasoning trajectories, with evaluation protocols for trajectory selection.

Undergraduate Researcher

McAuley Lab, UC San Diego

Advised by Prof. Julian McAuley and Yupeng Hou.

• Built an evaluation framework for benchmarking LLM text embeddings in retrieval and recommendation, including a public Amazon dataset with 2M+ training and 250K+ test instances.
• Benchmarked embeddings from SimCSE, E5-Mistral-7B, text-embedding-3-large, Qwen-7B, Gemini, and OpenAI across collaborative filtering, sequential recommendation, and product search; co-first-authored the resulting ACL 2026 paper.

Undergraduate Researcher

Shang Data Lab, UC San Diego

Advised by Prof. Jingbo Shang and Letian Peng.

• Refined weakly supervised text classification, reaching 93% peak accuracy on AG News (vs. 78% baseline) via KL-divergence matching between token-level text and class-prompt distributions.
• Implemented the framework in PyTorch with BERT, RoBERTa, and Gemma across AG News and Rotten Tomatoes.

M.S.E. in Computer & Information Science

University of Pennsylvania

• GPA: 3.95 / 4.0
• Relevant coursework: Statistical NLP, Deep Learning, Deep Generative Modeling, Reinforcement Learning, Machine Learning, Recommender Systems & Web Mining, Probability & Statistics

B.S. in Mathematics–Computer Science

University of California, San Diego

• GPA: 3.9 / 4.0 · Provost Honors

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

ACL 2026 Main Conference (Oral)

Yupeng Hou†, Jiacheng Li†, Xiangjun Fu†, Zhankui He, An Yan, Xiusi Chen, Julian McAuley. († equal contribution)

Latent Reasoning with Normalizing Flows

COLM 2026 Workshop on Efficient Reasoning

Guancheng Tu†, Xiangjun Fu†, Suhao Yu, Yao Tang, Haoqiang Kang, Lianhui Qin, Yizhe Zhang, Jiatao Gu. († equal contribution, order by coin toss)

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

Under review · arXiv:2510.26001

Xinhua Wang, Xiangjun Fu, Caibo Feng, Chunxiao Liu.