Institution profile

Lingnan University

Academic institutionasia · hk
Official website
Research library190linked papers
Opportunities0open roles
Selected work

Representative Papers

AI Survival Stories: a Taxonomic Analysis of AI Existential Risk

Jan 14, 2026

This study addresses whether advanced artificial intelligence constitutes an existential threat to humanity and systematically analyzes plausible pathways for human survival. Building upon the premises that “AI will become extremely powerful” and “if AI becomes extremely powerful, it will destroy humanity,” the work proposes the first comprehensive typology of AI existential scenarios, translating abstract possibilities of survival into concrete, analyzable trajectories. Through logical analysis, philosophical argumentation, and risk modeling, the paper identifies four primary classes of survival narratives, elucidates their key challenges and policy implications, and offers a preliminary quantitative foundation for assessing the probability of AI-induced existential catastrophe (P(doom)).

8 citations1 influentialRead paper

iExam: A Novel Online Exam Monitoring and Analysis System Based on Face Detection and Recognition

Jun 27, 2022arXiv.org

The COVID-19 pandemic accelerated the adoption of online proctoring via platforms like Zoom, yet real-time multi-stream video monitoring suffers from low efficiency and poor detection of anomalous behaviors. To address this, we propose the first lightweight intelligent proctoring system tailored for Zoom-based examinations. Our method introduces a novel automatic face ground-truth labeling technique that fuses dynamic name tags (positioned at the top-left corner of each participant’s window) with enhanced OCR. We further design a streaming video processing framework integrating: (i) lightweight YOLOv5 for real-time face detection; (ii) FaceNet embeddings with cosine similarity for identity verification; and (iii) PaddleOCR with dynamic coordinate mapping for robust text localization. The system achieves 90.4% real-time face detection accuracy and 98.4% accuracy in post-exam anomaly detection—including off-screen behavior, face turning, and impersonation—while maintaining sub-300 ms latency on standard faculty PCs. Source code is publicly available.

4 citationsRead paper

FACT: A Forensic Agent with Compiled Tool-Use Trajectories for AI-Generated Image Detection

Sep 05, 2026

AI-generated image detection is increasingly open-world: new image generators produce highly realistic images that make visual artifacts harder to identify. Existing detectors usually rely on a fixed set of forensic cues, so a detector that works well for one generator family may fail on another. We introduce FACT (Forensic Agent with Compiled Tool-use Trajectories), which learns an image-conditioned tool-use policy for forensic analysis. Instead of applying a fixed detector, FACT decides which forensic tools to call, interprets the returned evidence, and stops when sufficient evidence has been collected. FACT follows an Evolve--Distill--Refine pipeline: it evolves an execution-verified forensic skill, compiles the skill into action--observation tool-use trajectories, distills them into a compact agent, and refines the policy with cost-aware GRPO. Across two internal and four public benchmarks, FACT achieves the best performance among all compared methods, including on recent unseen generators, deepfakes, and manipulated images.

0 citationsRead paper
Recent publications

Latest Papers

FACT: A Forensic Agent with Compiled Tool-Use Trajectories for AI-Generated Image Detection

Sep 05, 2026

AI-generated image detection is increasingly open-world: new image generators produce highly realistic images that make visual artifacts harder to identify. Existing detectors usually rely on a fixed set of forensic cues, so a detector that works well for one generator family may fail on another. We introduce FACT (Forensic Agent with Compiled Tool-use Trajectories), which learns an image-conditioned tool-use policy for forensic analysis. Instead of applying a fixed detector, FACT decides which forensic tools to call, interprets the returned evidence, and stops when sufficient evidence has been collected. FACT follows an Evolve--Distill--Refine pipeline: it evolves an execution-verified forensic skill, compiles the skill into action--observation tool-use trajectories, distills them into a compact agent, and refines the policy with cost-aware GRPO. Across two internal and four public benchmarks, FACT achieves the best performance among all compared methods, including on recent unseen generators, deepfakes, and manipulated images.

0 citationsRead paper