PhD Researcher · UTS CybeR Lab · Trustworthy AI

Delong Li

I study machine unlearning, agentic AI security, and privacy-aware inference, with a focus on making adaptive AI systems more auditable, governable, and empirically trustworthy.

Selected papers Scholar profile

Research agenda and technical focus.

I work on how AI systems can be updated, forgotten, audited, and deployed under real privacy, security, and governance constraints.

I am currently a PhD student at the University of Technology Sydney and a member of UTS CybeR Lab. My research connects machine unlearning, LLM security, federated and edge inference, evaluation methodology, and agentic AI governance, with an emphasis on systems that can be verified and responsibly operated.

Machine Unlearning AI Security Federated Learning Privacy-Aware Inference LLM Agents Auditable AI

Selected publications.

Recent papers across agentic systems, auditable LLM updates, privacy-aware edge-cloud inference, and machine unlearning.

2026 arXiv preprint

SoK: Agentic Skills -- Beyond Tool Use in LLM Agents

Yanna Jiang, Delong Li, Haiyu Deng, Baihe Ma, Xu Wang, Qin Wang, Guangsheng Yu

A systematization of agentic skills across their lifecycle, with taxonomies for skill packaging, representation, execution scope, evaluation, and governance risk.

Visual page arXiv
2026 Electronics

AuditableLLM: A Hash-Chain-Backed, Compliance-Aware Auditable Framework for Large Language Models

Delong Li, Guangsheng Yu, Xu Wang, Bin Liang

A lightweight audit layer for LLM adaptation and unlearning that records tamper-evident update trails while supporting third-party verification.

Publisher
2026 Journal of Cloud Computing

Semantic Neighbor Swapping for Privacy-Aware Edge-Cloud Inference

Delong Li, Baihe Ma, Yanna Jiang, Chen Li, Xuelei Qi, Xu Wang, Feifan Wang, Bin Liang, Guangsheng Yu

A post-hoc embedding replacement method that reduces identity leakage in edge-cloud inference while preserving task utility under fixed communication cost.

Publisher
2025 SSRN preprint

A Survey of LoRA-based Machine Unlearning for LLMs: Methods, Taxonomy, and Evaluation

Delong Li, Guangsheng Yu, Xu Wang, Yanna Jiang, Wencheng Yang, Bin Liang, Wei Ni

A survey of adapter-centric machine unlearning for LLMs, organizing methods through a unified adapter-space taxonomy and evaluation view.

Visual page SSRN

Projects and research artifacts.

Reusable research tooling and public artifacts that package methods and workflows for broader reuse.

Codex skill

Paper Visualization

A reusable Codex skill for turning paper PDFs, preprint pages, or source workspaces into readable static paper visualizations. The SoK page is maintained as a live example of the skill output.

GitHub repo Live example

Get in touch.

I am open to research discussions and collaborations around LLM security, machine unlearning, privacy, and auditable AI systems.