Spark (Wen-Kwang) Tsao, Ph.D.
Senior AI Safety Research Lead & Architect AI Safety β’ Reinforcement Learning β’ Adversarial Robustness β’ Red Teaming
Researcher exploring AI safety through production systems and personal experiments. I build safety-critical ML systems by day and explore new ideas through prototypes and experiments.
π Published Research
First Author
- Mapping Smarter, Not Harder: A Test-Time Reinforcement Learning Agent That Improves Without Labels or Model Updates β EMNLP 2025 (Industrial Track) Novel RL agent achieving 93.94% accuracy and reducing expert review workload by 85% through test-time learning without retraining. arXiv:2510.14900
Team Lead / Last Author
- Primus: A Pioneering Collection of Open-Source Datasets for Cybersecurity LLM Training β 2025 Open-source cybersecurity dataset suite (pretraining, fine-tuning, reasoning distillation) yielding 15.9% improvement in aggregate benchmarks and 15.8% gain in CISSP certification.
Additional Publications
- Multi-Agent Reasoning with Large Language Models for Effective Corporate Planning β IEEE CSCI 2023
π’ Professional Experience
Head of AI Lab @ Trend Micro (2008 β Present)
15+ years building safety-critical ML systems deployed globally. Founded and led AI Lab, delivering production AI safety systems from research to deployment:
- ML-based threat detection systems (2016)
- GenAI security copilot platforms (2023)
- Red-teaming pipelines and oversight systems at scale
- 20+ U.S. patents in AI-driven security systems
π Recognition
- 20+ U.S. Patents in AI security and detection systems
- 418 citations (h-index 10) β’ Google Scholar
- EMNLP 2025 Industrial Track first-author publication
π§ͺ GitHub: Exploration Lab
This GitHub hosts my experiments and explorationsβtoys for learning, prototypes for testing ideas, and explorations in AI safety, RL, and adversarial testing.
Selected themes:
- LLM agent evaluation and reasoning systems
- Red-team simulations and adversarial robustness
- Reinforcement learning for scalable oversight
π« Contact
Email: linhots.tsao@gmail.com LinkedIn: linkedin.com/in/spark-tsao Google Scholar: scholar.google.com/citations