Experience

Principal Researcher — M365 Core, Microsoft

Dates: August 2024–Present

Lead applied research and development for M365 BizChat / Copilot, the Researcher agent, and related AI assistants, with a focus on LLM-based reasoning over enterprise knowledge, deep search, evaluation, and trustworthy AI and security readiness.

• Lead applied research on AI-assistant reasoning quality, model selection, deep search, and trustworthy AI and security readiness in product-facing settings.
• Built offline evaluation and replay infrastructure for production reasoning traces, enabling systematic diagnosis of answer quality, citation quality, search-grounded reasoning, and model-selection quality.
• Advance the Researcher agent, deep search, and multi-turn conversation quality.
• Continue research on AI security, including LLM safety, agent security, model tamper detection and IP protection, and adversarial robustness.
• Contribute leadership to Microsoft's Secure Future Initiative (SFI) and security learning community.

Principal Researcher — Bing Core Search, Microsoft

Dates: February 2023–August 2024

Led product-facing research in multi-vector retrieval, prompt optimization, document understanding, retrieval measurement, and Web recommendation for Bing Core Search.

• Co-designed a flexible multi-vector retrieval system (XTR + ColBERT) as a new retrieval mechanism to replace sparse+dense retrieval.
• Designed prompt-optimization methods used across multiple Bing teams, including AMPO and StraGo (EMNLP 2024).
• Built document-understanding pipelines using large language models to generate synthetic queries and click signals for fresh and long-tail Web documents.
• Continued research on AI security, including LLM safety, agent security, model tamper detection and IP protection, and adversarial robustness.

Researcher → Lead Researcher → Principal Researcher — Microsoft Research Asia

Dates: December 2001–February 2023

Led a long-running research program in AI security, AI applications in security, trustworthy computing, privacy-preserving computation, content protection, and multimedia systems.

• Conducted research in responsible and trustworthy AI, AI/ML security, phishing and account protection, authentication, CAPTCHA, scalable-media encryption and DRM, applied cryptography, data embedding, visualization, and human-computation systems.
• Transferred major technologies including defense against human-operated ransomware, malicious PowerShell script detection, phishing detection, CAPTCHA, fraud detection, Bing index compression, and mobile integrity checking.
• Built collaborations across Microsoft Research Asia (MSRA), Microsoft Research Redmond, Microsoft product teams, university partners in Asia, the U.S., and the U.K., and external research communities, taking ideas through prototypes, evaluation, transfer, and long-running product use.

Selected Project Themes

• AI security and AI applications in security: backdoor detection and data cleansing, evasion-robust detectors, and behavior-based account protection.
• Human interactive proofs and authentication: image, manipulable, and overlay CAPTCHAs, audio HIPs, graphical-password security analysis, and device-assisted authentication.
• Phishing, fraud, and account abuse: ML and content-based detection, evasion-resistant logo recognition, and explainable anti-phishing models under strict privacy constraints.
• Multimedia security and content protection: scalable, syntax-compliant encryption and DRM for MPEG-4 and JPEG2000, and data embedding into images, charts, and slides (Chartem, ChartStamp).
• Privacy-preserving computation and cloud reliability: decompression and analytics in the homomorphic-encryption domain, encrypted-video motion detection, and unsupervised anomaly and outage detection for Azure.

Lead Scientist and Co-Founder — Cognicity, Inc.

Location: Edina, MN   Dates: August 1997–September 2001

Co-founded a multimedia watermarking and data-embedding startup based on Ph.D. research at the University of Minnesota. Led research and development for audio, image, and video watermarking, fingerprinting, copyright protection, access control, tracking, and linking products.

• Provided technical leadership for new algorithms and core technologies, designed the product architecture, and turned research prototypes into commercial products.
• Led development of AudioKey (retail audio watermarking), AudioKey MP3 tMark and Cognicity tMark (transactional, download-time watermarking), and TurnKey (per-CD unique-ID watermarking at burn time).
• Contributed to music-industry copyright-protection efforts, including MUSE, the 4C Entity DVD-Audio proposal, and SDMI Phase 2 screening, and built a covert-communication product for a U.S. government customer.

Academic Appointments and Teaching

• Adjunct Professor, Huazhong University of Science and Technology, 2019–2023.
• Teaching
  • Course Instructor, Data Mining and Business Intelligence, School of Software, Beihang University, 2012.
  • Course Instructor, Data Warehouse and Data Mining, School of Software, Beihang University, 2010.
  • Course Instructor, Information Systems Management, School of Software, Beihang University, 2009.
  • Teaching Assistant, Department of Physics, University of Minnesota, 1989–1992; taught undergraduate physics laboratories.