Dongxiao Zhu
Wayne State University, MI·Computer Science
AI language tools and dating apps raise linked problems of safety and privacy. Large language models can memorize private or harmful content, and online interactions can lead to real-world sexual harm.
- Fields
Computer Science
Social Science
Public Health & Epidemiology
- Topics
large language models
machine unlearning
LLM unlearning
- Grants
- 2 known grants, through Sep 2029
What they work on
- Develop optimization algorithms that make large language model unlearning, removing specific data influences without retraining, more effective, robust and efficient.
- Use interpretability methods to identify which internal LLM weights drive memorization and adjust them without losing the model's emergent abilities.
- Examine how watermarking and coreset selection affect unlearning outcomes, including methods for imperfect or proxy sets of data to be forgotten.
- Apply unlearning to conversational risk assessment in online dating, letting LLMs evaluate behavioral risks while erasing personal information.
- Design, implement and evaluate a privacy-preserving data donation app through which online daters report experiences related to sexual consent.
- Build a framework of online-to-offline sexual violence through longitudinal data donation and recurring interviews with cohorts of online daters.
More about the research
This research builds computational methods to make AI models forget unwanted information. It also studies how online communication shapes sexual consent, with online dating as a shared application.
Why it matters
Unlearning methods could let companies remove private user data, copyrighted text or unsafe instructions from deployed AI systems without costly retraining. That would strengthen privacy in healthcare, education and cybersecurity tools. The dating research targets sexual violence, a major public health problem. It could inform prevention tools that do not depend on perpetrators or victims recognizing harm as it happens.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Lu Lin
Pennsylvania State University, PA·Computer Science
Identify critical LLM safety concerns, gaps in current evaluation and infrastructure needs through structured surveys and interviews with experts across computing disciplines.
Xin Zhang
San Diego State University, CA·Computer Science
Establish diagnostic theory tying spatial-temporal tokenization, cross-modal representation alignment, and distributional drift to the conditions under which LLM-based urban forecasters generalize or break down, producing measurable criteria usable before deployment.
Jie Yan
Bowie State University, MD·Computer Science
Investigate the sources of hallucination — fabricated or inaccurate output — when large language models are applied to digital forensic casework, and develop techniques that reduce it enough for results to align with evidentiary standards.
The school
Wayne State University: 21 labs on PI Pages, $292M a year on research, 194 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.