Jie Yan
Bowie State University, MD·Computer Science
Digital evidence — what is recoverable from phones, laptops and internet-connected gadgets — increasingly decides legal cases, yet the tools used to analyze it are getting harder to check. Work in this group sits at the meeting point of artificial intelligence, digital forensics and cybersecurity, asking two linked questions: can AI systems be made reliable enough for evidence-grade use, and can AI spot security flaws in connected devices before attackers do.
- Fields
Computer Science
Engineering
Social Science
- Topics
digital forensics
large language models
hallucination mitigation
- Grants
- 2 known grants, through Aug 2029
What they work on
- 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.
- Make LLM forensic output interpretable using explainable-AI techniques, and design and validate performance metrics plus visualization-based representations that systematically evaluate model reliability.
- Build and test a virtual reality training environment in which users interact with forensic scenarios and evaluate LLM-generated evidence in context.
- Use deep learning to recover hidden content and encrypted messages from Internet of Things devices in a forensically sound manner, assessing where personal data privacy is exposed.
- Analyze IoT application source code for vulnerability classes including buffer overflow, integer overflow and carriage-return/line-feed injection, and reconstruct attack scenes from forensic evidence to reveal system vulnerabilities.
More about the research
One project targets large language models (LLMs), the text-generating AI systems behind chat assistants, as forensic aids. Their main liability is hallucination: confidently stated but fabricated output. The project investigates where hallucinations come from in forensic settings, develops techniques to reduce fabricated or inaccurate answers so results can meet evidentiary standards, and applies explainable-AI methods so an investigator can see why the model said what it said. The team will design and validate performance metrics and visualization-based representations that let LLM reliability be evaluated systematically rather than judged case by case. Alongside this, the PIs build a virtual reality (VR) training environment in which users walk through forensic scenarios and judge LLM-generated evidence in context, bringing human-computer interaction and user-centered design into the evaluation.
Why it matters
Investigators, courts and law enforcement are starting to lean on AI tools whose output can be wrong in ways that are hard to detect; methods for measuring and explaining that reliability could keep unreliable AI conclusions out of legal decisions. The IoT work could help device makers and users close privacy and code-level security holes before attackers exploit them, with techniques that extend to cybersecurity and healthcare systems.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Matthew Wright
Rochester Institute of Technology, NY·Computer Science
Study how users evaluate suspect digital content through qualitative interviews and user studies, and use the findings to integrate multiple AI-content forensic tools into one navigable platform.
Sudhir Trivedi
Southern University and A & M College, LA·Computer Science
Identify and cluster Internet-connected EV management stations at scale using fingerprinting that combines automated web scraping, recursive unsupervised learning and pattern matching, and retrieve their firmware and software for analysis.
Dongxiao Zhu
Wayne State University, MI·Computer Science
Develop optimization algorithms that make large language model unlearning, removing specific data influences without retraining, more effective, robust and efficient.
The school
Bowie State University: 1 labs on PI Pages, $4M a year on research, 10 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.