CV
PhD Candidate, University of Michigan · NLP & LLM Researcher
yinuoxu@umich.edu · Ann Arbor, MI · LinkedIn · GitHub · Google Scholar
Education
University of Michigan, Ann Arbor 2023 – Present
Ph.D. in Information · Advisor: David Jurgens
University of California, Berkeley 2023
B.A. in Data Science & Statistics
Awards, Grants & Leadership
- Co-Organizer, Personalization Without Prejudice (PwP) workshop proposal under review, ACL 2026
- Organizer, Michigan NLP Reading Group, 2025–2026
- Organizer, NLP@Michigan Day 2026, 2025–2026
- Science Communication Fellow (selected as 1 of 20 fellows university-wide), 2023–2024
- University of Michigan School of Information Merit Fellowship, 2023
Publications
- EMNLP 2026 Xu, Y., & Jurgens, D. (2026). Beyond consensus: Perspectivist modeling and evaluation of annotator disagreement in NLP.
- ACL 2026 ยท Oral Xu, Y., Derricks, V., Earl, A., & Jurgens, D. (2026). Modeling annotator disagreement with demographic-aware experts and synthetic perspectives.
- NAACL 2025 Xu, Y., Chen, H., Rakshit, S., et al., & Jurgens, D. (2025). Causally modeling the linguistic and social factors that predict email response. (Equal contribution)
- JEDM 2025 Borchers, C., Xu, Y., & Pardos, Z. A. (2025). Workload overload? Late enrollment leads to course dropout. Journal of Educational Data Mining, 17(1), 126–156.
- EDM 2024 Borchers, C., Xu, Y., & Pardos, Z. A. (2024). Are you an early dropper or late shopper? Mining enrollment transaction data to study procrastination in higher education.
- LAK 2024 Xu, Y., & Pardos, Z. A. (2024). Extracting course similarity signal using subword embeddings.
- EDM 2023 Xu, Y., & Pardos, Z. A. (2023). Mining detailed course transaction records for semantic information.
Relevant Experience
Research & Development Intern, Kitware May 2026 – Present
Project: Personalized LLM Steering from Inferred Value Latents
- Designed an end-to-end pipeline that infers individual users' latent value profiles from a few survey responses, with a Bayesian active-learning module adaptively selecting the most informative elicitation questions.
- Built a variational information-bottleneck encoder compressing responses into a low-dimensional value embedding with demographic priors for cold-start, then a hypernetwork amortizing the latent-to-steering-vector map so new users require no optimization, gated by posterior uncertainty.
- Evaluated against SOTA steering methods across four value datasets: +0.22 user-specific accuracy over zero-shot, outperforming existing steering methods.
- Ran interpretability analysis recovering a binding-vs-individualizing moral foundations axis in the learned steering vectors.
Lead Researcher, University of Michigan School of Information Aug 2023 – Present
Project 1: Modeling Annotator Disagreement with Demographic-Aware Experts
- Designed Demographic-Aware Mixture-of-Experts, an interpretable architecture with demographic-aware routing to capture structured variation in annotation behavior and intersectional perspectives.
- Achieved state-of-the-art subgroup-level performance across five benchmark datasets, particularly under data imbalance or sparse subgroup representation.
- Evaluated LLM annotation reliability through zero-shot, few-shot, and LoRA-finetuned experiments, measuring alignment between persona-prompted LLM ratings and human judgments.
- Developed data-efficient pipelines blending real and LLM-generated synthetic annotations with alignment-weighted loss; implemented multi-GPU training workflows for model scaling and reproducibility; paper accepted at ACL 2026 (arXiv:2508.02853).
Project 2: B-HAP: Bayesian Hierarchical Adapter Personas for Situated Judgment Modeling
- Designing a hypernetwork-based framework generating parameter-efficient LoRA adapters from task, data, and annotator-feature interactions, with a three-level Bayesian hierarchy (population, subpopulation, individual) to align model behavior with group norms.
- Leading evaluation across a 140K-instance benchmark spanning normative, pragmatic, and preference-judgment tasks, benchmarked against Jury Learning, Mixture-of-Personas, and sociodemographic prompting baselines.
Lead Researcher, UMich School of Information & Social Psychology Aug 2023 – Present
Project: Measuring Racial Disparities in Doctor-Patient Conversations
- Built a large-scale annotation pipeline to evaluate doctor communication quality using a diverse pool of crowdworkers; designed and deployed a custom annotation interface to ensure demographic balance and inter-rater reliability.
- Developed a zero-shot LLaMA-70B annotation pipeline to label doctor behaviors (respect, formality) with prompt-engineered templates, validating alignment with human ratings.
- Designed and implemented a causal mediation model linking race → doctor language → respect perception → patient trust, revealing differential behavior–trust pathways across demographic groups.
- Paper in preparation for PNAS.
Graduate Research Assistant, UMich School of Information March 2024
Project: Causal Modeling of Linguistic and Social Factors in Email Response (NAACL 2025)
- Co-led the development of causal inference models analyzing linguistic and social predictors of email responsiveness, combining structural equation modeling with LLM-based feature extraction.
- Conducted large-scale language and behavior modeling using Transformer-based encoders and mixed-effects regression to uncover latent causal drivers of social reciprocity.
- Presented poster at NAACL 2025; paper accepted in the main conference (aclanthology.org/2025.naacl-long.594).
NLP Intern, Evisort (acquired by Workday) 2021
- Developed algorithms to redact sensitive information in legal contracts using the Google DLP API and custom regex/XML pipelines.
- Trained and deployed LightGBM and bi-LSTM NER models in TensorFlow for contract field extraction on a SaaS platform used by enterprise clients.
- Designed PowerBI dashboards and PostgreSQL queries to analyze document metadata and usage patterns from AWS data sources.
Teaching Experience
- Becoming a Data Scientist (SI 568), Graduate Student Instructor, University of Michigan, Winter 2026
- Applied Machine Learning (SI 670), Graduate Student Instructor, University of Michigan, Fall 2025
- Natural Language Processing (Info 159/259), Teaching Assistant, UC Berkeley, Spring 2023
- Principles and Techniques of Data Science (Data 100/200), Undergraduate Student Instructor, UC Berkeley, Fall 2022
- Principles and Techniques of Data Science (Data 100/200), Undergraduate Student Instructor, UC Berkeley, Summer 2022
- Technology and Social Impact (EECS 198), Lead Facilitator, UC Berkeley, Spring 2023
Professional Service
- Program Committee Member, International Conference on Computational Social Science (IC2S2), 2026
- Program Committee Member, Workshop on NLP and Computational Social Science (at ACL), 2026
Technical Skills
Python · PyTorch · Transformers · vLLM · Optuna · ModernBERT · LoRA Fine-Tuning · Mixture-of-Experts · Distributed / Multi-GPU Training (PyTorch DDP) · Reinforcement Learning · Synthetic Data Generation · Causal Inference · Predictive & Statistical Modeling · LLM Evaluation & Alignment · Instruction-Tuning · Data Efficiency · Prompt Engineering · Data Pipeline Design · Bayesian Modeling