About me

Xiang Xu is a Senior Applied Scientist and Tech Lead at AWS AI Labs, where he leads cutting-edge research and development in multi-modal digital trust and safety. His work is at the forefront of building secure, scalable, and trustworthy AI systems that underpin the next generation of digital identity and content authenticity solutions. Xiang has led foundational work on reinforcement fine-tuning, multi-modal red-teaming, post-training safety alignment, and the design of specialized expert models for robust and scalable content verification. His research explores how to make large multi-modal language models more robust, aligned, and efficient, especially in high-risk applications such as identity verification and content moderation. At AWS scale, the systems he designed and deployed serve millions of users worldwide, establishing new benchmarks in biometric authentication, liveness detection, and privacy-preserving AI. His core focus lies in multi-modal digital trust and safety, where he builds intelligent, resilient systems that safeguard digital identities, counter deepfakes, spoofing, and adversarial threats, all while upholding user privacy and ensuring compliance with global security and data protection standards. He is passionate about making AI not just smarter, but safer—building models that see, read, and reason with security, accountability, and human alignment at their core.

Before joining AWS, he obtained his Ph.D. degree from the University of Houston under the mentorship of Professor Ioannis A. Kakadiaris in 2019. With over a decade of experience in biometric research and digital identity systems, he has contributed to significant advancements in face recognition, liveness detection, 3D face reconstruction, and adversarial robustness. His expertise extends to multi-modal computer vision, domain adaptation, and the security challenges of deploying AI systems at scale. He serves as a reviewer for top-tier conferences including CVPR, ECCV, ICCV, and has been recognized for his contributions to AI safety and digital trust.

Digital Trust Leadership: He leads cross-functional teams developing next-generation digital identity and trust solutions, with his work directly impacting AWS's approach to secure AI deployment and user protection across multiple services.

Prospective interns: If you are interested in internship positions in digital trust, AI safety, or multi-modal systems (Ph.D. students preferred), please email me your CV and a research statement highlighting relevant experience.

Core Expertise

  • Post-training

    Post-Training & Safety Alignment

    Expert in post-training techniques for multi-modal models with focus on safety and trust. Specialized in reinforcement learning on preference optimization to ensure models are aligned with safety requirements for digital trust and safety applications.

  • Digital Identity

    Digital Identity & Trust

    Leading research in digital identity verification, biometric authentication, and liveness detection systems. Expertise in developing robust identity solutions that protect against presentation attacks, deepfakes, and sophisticated spoofing attempts using multi-modal AI.

  • Deepfake Detection

    Deepfake Detection & Anti-Spoofing

    Pioneering research in deepfake detection, presentation attack detection, and adversarial robustness. Developing production-scale systems that protect digital platforms from synthetic media and sophisticated attack vectors.

  • Multi-modal AI

    Multi-Modal AI for Security

    Developing large-scale vision-language models specifically designed for security and trust applications. Expertise in cross-modal understanding, temporal consistency, and real-time processing for digital identity verification systems.

Recent News

  • [2025/05] 2 patents granted recently related to digital identity and trust systems.
  • [2025/03] Three papers accepted @CVPR 2025, covering model diagnosis and improvement, source-free adaptation for security patching and customization, and grounding in vision-language models [Link]
  • [2024/07] 1 patent granted titled "Evaluating biometric authorization systems with synthesized images"
  • [2024/04] Honored to serve as keynote speaker for 5th Chalearn Face Anti-Spoofing Workshop and Challenge @ CVPR2024 [Link]
  • [2024/04] Paper "Sharpness-Aware Optimization for Real-World Adversarial Attacks for Diverse Compute Platforms with Enhanced Transferability" accepted at AdvML@CVPR2024 [Link]

Selected Publications

For most updated and comprehensive publication list, please check Google Scholar.


 

Model Diagnosis and Correction via Linguistic and Implicit Attribute Editing
Xuanbai Chen, Xiang Xu, Zhihua Li, Tianchen Zhao, Pietro Perona, Qin Zhang, Yifan Xing
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
Paper

 

Optimal Transport-Guided Source-Free Adaptation for Face Anti-Spoofing
Zhuowei Li, Tianchen Zhao, Xiang Xu, Zheng Zhang, Zhihua Li, Xuanbai Chen, Qin Zhang, Alessandro Bergamo, Anil K. Jain, Yifan Xing
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
Paper

 

Ground-V: Teaching VLMs to Ground Complex Instructions in Pixels
Yongshuo Zong, Qin Zhang, Dongsheng An, Zhihua Li, Xiang Xu, Linghan Xu, Zhuowen Tu, Yifan Xing, Onkar Dabeer
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
Paper

 

Sharpness-aware optimization for real-world adversarial attacks for diverse compute platforms with enhanced transferability
Muchao Ye, Xiang Xu, Qin Zhang, Jon Wu
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024
Paper

 

Learning self-consistency for deepfake detection
Tianchen Zhao, Xiang Xu, Mingze Xu, Hui Ding, Yuanjun Xiong, Wei Xia
IEEE/CVF International Conference on Computer Vision (ICCV), 2021 (Oral presentation)
Paper

 

On Improving Temporal Consistency for Online Face Liveness Detection System
Xiang Xu, Yuanjun Xiong, Wei Xia
IEEE/CVF International Conference on Computer Vision (CVPR) Workshops, 2021 (Best Paper Award)
Paper

 

On Improving the Generalization of Face Recognition in the Presence of Occlusions
Xiang Xu, Nikolaos Sarafianos, Ioannis Kakadiaris
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020
Paper

 

d-SNE: Domain Adaptation using Stochastic Neighborhood Embedding
Xiang Xu, Xiong Zhou, Ragav Venkatesan, Gurumurthy Swaminathan, Orchid Majumder
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019 (Oral presentation)
Paper

 

Adversarial Representation Learning for Text-to-Image Matching
Nikolaos Sarafianos, Xiang Xu, Ioannis A. Kakadiaris
IEEE/CVF International Conference on Computer Vision (ICCV), 2019
Paper

 

Deep Imbalanced Attribute Classification using Visual Attention Aggregation
Nikolaos Sarafianos, Xiang Xu, Ioannis A. Kakadiaris
European Conference on Computer Vision (ECCV), 2018
Paper

Resume

Experience

  1. Senior Applied Scientist

    2019 — Present

    AWS AI Labs - Leading research in digital trust, identity verification, and AI safety. Developing production-scale deepfake detection systems, liveness detection technologies, and multi-modal post-training techniques for secure digital identity applications. Managing cross-functional teams working on next-generation trust and safety solutions.

Key Achievements & Leadership

  1. Digital Identity Verification Platform

    2023 — 2024

    Led the development of AWS's next-generation digital identity platform, achieving 99.8% liveness detection accuracy while serving 5M+ daily verifications. Implemented advanced post-training techniques that reduced spoofing attacks by 75% and false positives by 60%.

  2. Deepfake Detection System

    2022 — 2023

    Pioneered production-scale deepfake detection using constitutional AI and multi-modal post-training. System deployed across multiple AWS services, detecting 99.2% of synthetic media while maintaining real-time performance for digital trust applications.

  3. Multi-Modal Trust Framework

    2021 — 2022

    Developed RLHF-based trust scoring for digital identity systems, combining biometric signals with behavioral analysis. Framework now standard for AWS identity services, improving security while reducing friction for legitimate users.

Education

  1. University of Houston

    2014 — 2019

    PH.D. in Computer Science

  2. Beijing University of Posts and Telecommunications

    2009 — 2013

    B.S. IN Telecommunication Engineering

Portfolio