Yixin Liu (刘奕鑫)

News & Highlights

• [Seeking Position] I am actively looking for full-time research scientist/engineer positions starting May 2026. My expertise spans LLMs, reinforcement fine-tuning (Group Relative Policy Optimization/GRPO), synthetic content detection, audio language models, AI safety, content protection, watermarking, and trustworthy ML. Please reach out if you have opportunities!
• [2025.08] Developed explainable audio synthetic content detection system at Dolby Labs using LLMs with reinforcement learning, achieving better explainability, generalizability, and effectiveness.
• [2025.01] Our XAttnMark is accepted by ICML'25! State-of-the-art neural audio watermarking achieving joint detection and attribution. [Virtual Poster]
• [2024.09] Started research internship at Dolby Labs, working on audio watermarking and content protection for Universal Music Group.
• [2024.07] Invited Talk at Microsoft ASG Research Day on "Adversarial Perturbation in Personalized Diffusion Models".
• [2024.06] Presented MetaCloak as CVPR'24 Oral - watch the talk!
• [2024.05] Our FViT is accepted by ICML'24 as Spotlight!
• [2024.05] Completed internship at Samsung Research America on Graph-based RAG for log analysis.
• [2023.12] Our Stable Unlearnable Example (SEM) is accepted by AAAI'24! Achieving 3.91× speedup with improved protection efficacy.

Research Interest

I am broadly interested in content protection in the era of generative AI and trustworthy ML. My research focuses on the intersection between powerful generative AI and copyright protection, developing data-centric approaches to safeguard user data from unauthorized exploitation and provide robust source verification.

• Proactive UGC Protection [MetaCloak (CVPR'24 Oral), DiffShortcut, SEM (AAAI'24), EditShield (ECCV'24), MUE (ICML'24 Workshop), GraphCloak, Linear Solver Analysis]: User-generated content faces unprecedented threats from unauthorized AI training. By exploiting the fundamental vulnerability that neural networks are not robust to small input perturbations, we develop protective mechanisms that prevent unauthorized model training while preserving data utility for legitimate use.
• AIGC Watermarking and Detection [XAttnMark (ICML'25), TextMarker]: As generative models blur the boundary between real and fake content, robust authentication becomes critical. We develop watermarking techniques that enable AI content attribution and source verification across modalities, including state-of-the-art neural audio watermarking with joint detection and attribution capabilities.
• Robust Explainable AI [SEAT (AAAI'23 Oral), FViTs (ICML'24 Spotlight)]: The exploration of adversarial learning and explainability mechanisms enhances our understanding of model vulnerabilities and interpretability. SEAT addresses attention mechanism instability, providing stable and explainable attention for NLP tasks. FViTs develops faithful vision transformers through denoised diffusion smoothing, ensuring robust attention maps under adversarial attacks. These works advance our understanding of adversarial learning, robustness red-teaming, and robust explainable AI foundations.

Professional Experience

• Dolby Labs - Research Intern (Sep 2024 - Apr 2025, May 2025 - Aug 2025)
  Working on robust audio watermarking for content protection with Universal Music Group. Developed XAttnMark achieving state-of-the-art detection and attribution performance. Extended work to explainable audio synthetic content detection using LLMs with reinforcement learning (using VERL), focusing on watermark-free scenarios.
• Samsung Research America - Research Intern (May 2024 - Aug 2024)
  Developed graph-based RAG system for log analysis, achieving +16 comprehensiveness score improvement. Also worked on DiffShortcut for defending protective perturbations in diffusion models.
• Samsung Research America - Research Intern (May 2023 - Nov 2023)
  Proposed efficient defensive perturbation generation methods for data protection against diffusion models, resulting in MetaCloak (CVPR'24 Oral) and GraphCloak for graph data protection.
• Lehigh University - Teaching Assistant
  CSE 017 Java Programming (Spring 2023), CSE 007 Python Programming (Spring 2024)

Invited Talks & Presentations

• ICML 2025 Poster - "XAttnMark: Learning Robust Audio Watermarking with Cross-Attention" [Virtual Poster]
• Dolby Lab Tech Summit - "Robust Audio Watermarking for the Music Industry" (June 2025) [Slides]
• Microsoft ASG Research Day - "Adversarial Perturbation in Personalized Diffusion" (invited by Dr. Tianyi Chen, July 2024) [Slides]
• CVPR 2024 Oral - "MetaCloak: Preventing Unauthorized T2I Diffusion Synthesis" (June 2024) [Video] [Slides]

Reviewer Service

NeurIPS'23'24, KDD'23'25, CVPR'24'25, ICML'24'25, ECCV'24 (Outstanding Reviewer), ICLR'25, ICASSP'25, IEEE TIP