Zijun Yu PhD student · McGill

Curriculum Vitae

Download PDF

PhD student in Mathematics and Statistics developing distribution-free methods for uncertainty quantification and risk control, with a current focus on the reliability, evaluation, and safety of LLMs and autonomous agents. Co-first-author papers at ICML 2026 and NeurIPS 2026, plus a NeurIPS workshop paper on agent safety and one manuscript under review.

Education

2025 – 2029 (exp.)
Ph.D. in Mathematics and Statistics Direct-entry
McGill University, Montréal, QC
Supervisors: Prof. Masoud Asgharian, Prof. Vahid Partovi Nia
Research: distribution-free uncertainty quantification and risk control; reliability and safety of LLMs and autonomous agents
2016 – 2020
B.Sc. Honours in Statistics and Computer Science
McGill University, Montréal, QC
First-Class Honours Degree

Awards & Scholarships

Publications

† Equal contribution.

Conference

  1. Yu Gu†, Zijun Yu†, Vahid Partovi Nia, Masoud Asgharian. "Pause and Reflect: Conformal Aggregation for Chain-of-Thought Reasoning." Conference on Neural Information Processing Systems (NeurIPS), 2026. arXiv ↗
  2. Yu Gu†, Zijun Yu†, Chi-Kuang Yeh, Xinyu Wang, Ziyang Song. "Severity-Controlled Prediction Sets for Medication Recommendation." Conference on Neural Information Processing Systems (NeurIPS), 2026.
  3. Pengqi Liu†, Zijun Yu†, Mouloud Belbahri, Arthur Charpentier, Masoud Asgharian, Jesse C. Cresswell. "Beyond Procedure: Substantive Fairness in Conformal Prediction." International Conference on Machine Learning (ICML), 2026. arXiv ↗ Code ↗

Workshop

  1. Zijun Yu†, Yu Gu†, Vahid Partovi Nia, Masoud Asgharian. "G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm." NeurIPS Workshop on SLMs for Agentic Systems (SLM-Agents), 2026. arXiv ↗

Under Review

  1. Zijun Yu†, Yu Gu†, Vahid Partovi Nia, Masoud Asgharian. "Conformal Prediction under Partial Verification." Under review, 2026. arXiv ↗

Presentations

  1. "G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm." Workshop presentation (upcoming). NeurIPS Workshop on SLMs for Agentic Systems (SLM-Agents), Paris, France, 2026.
  2. "Beyond Procedure: Substantive Fairness in Conformal Prediction." Poster presentation. International Conference on Machine Learning (ICML), Seoul, South Korea, 2026.

Service

Patents & Open-Source Software

Work Experience

2025 – 2026
Student Researcher Mitacs Internship
Layer 6 AI, Toronto
  • Conducted research on substantive fairness in conformal prediction, leading to "Beyond Procedure: Substantive Fairness in Conformal Prediction" (ICML 2026) and a filed patent application.
  • Designed and architected a modular codebase supporting large-scale experiment pipelines, enabling reproducible benchmarking across multiple fairness criteria, datasets, and model families; released open source.
  • Collaborated with research scientists on theoretical analysis and empirical validation.
2020 – 2024
Software Developer
OneDesk Inc., Montréal
  • Built a Retrieval-Augmented Generation chatbot pipeline (web crawler, vector database, LLM APIs) and added data analysis and visualization capabilities to the core product.
  • Diagnosed and resolved memory leaks, race conditions, and data-consistency failures across frontend and backend systems; applied time-series analysis to production logs to audit system performance and identify regressions.
  • Led and mentored a team of junior developers across multiple concurrent projects.

Technical Skills

ML PyTorch, Hugging Face Transformers, vLLM, llama.cpp
Languages Python, R, Java, TypeScript, SQL
Infrastructure Docker, PostgreSQL, Redis, Git, Proxmox VE