Xiaoqian Lu

About

I am an incoming Ph.D. student at the University of Ottawa (starting Sep 2026), where I will be advised by Prof. Yiwei Lu.

My research focuses on Trustworthy Machine Learning and LLM Safety. I am particularly interested in:

LLM Safety — understanding and mitigating vulnerabilities in LLM post-training (e.g., data poisoning attacks), investigating memorization behaviors and privacy risks, and ensuring safe and controllable recursive self-improvement (RSI) in AI systems
Test-Time Training — adaptive learning and model refinement at inference time

Publications

SSR: Similarity-Shift Refinement for Training-Free Object-Centric Masks

Xiaoqian Lu, Guangfu Guo

Preprint / arXiv

Object-Centric Learning Vision Transformer

DeltaFlow: Noise-Adaptive Bidirectional Gated Delta Networks for Embedded Language Flows

Guangfu Guo, Xiaoqian Lu, Linsey Pang, Weiran Yao, Haolin Chen, Kunpeng Liu, Long Cheng

Preprint / arXiv

Language Model Efficient Architecture

Sequential Data Poisoning in LLM Post-Training

Jack Sanderson, Yihan Wang, Xiaoqian Lu, Gautam Kamath, Yiwei Lu

Preprint / arXiv

Data Poisoning LLM

Beyond Static Visual Tokens: Structured Sequential Visual Chain-of-Thought Reasoning

Guangfu Guo, Xiaoqian Lu, Yue Feng, Mingming Sun

EMNLP 2026 Findings / arXiv

VLM Chain-of-Thought

Med-VRAgent: A Framework for Medical Visual Reasoning-Enhanced Agents

Guangfu Guo, Xiaoqian Lu, Yue Feng

EMNLP 2025 / arXiv

Medical AI Visual Reasoning

Structured Outputs Enable General-Purpose LLMs to be Medical Experts

Guangfu Guo, Kai Zhang, Bryan Hoo, Yujun Cai, Xiaoqian Lu, Nanyun Peng, Yiwei Wang

Preprint / arXiv

Medical AI LLM

Education

2026 – Present University of Ottawa, Ph.D.
2023 – 2025 University of Bristol, M.S.
2018 – 2020 Yokohama National University, Exchange
2016 – 2021 East China Normal University, B.S.