Yuansen Liu (刘源森)

Yuansen Liu (刘源森)

MComp (AI), National University of Singapore · Incoming PhD applicant

Yuansen Liu刘源森

About | 个人简介

I am Yuansen Liu (刘源森), an MComp (AI) student at the National University of Singapore (NUS) (expected graduation: November 2026). I am currently applying for PhD programs in trustworthy and multimodal AI.

我目前就读于新加坡国立大学（NUS）MComp AI项目（预计 2026 年 11 月毕业），正在申请博士，研究方向为可信大模型、多模态推理与医疗智能。

Previously, I received my B.Eng. in Software Engineering from Beihang University (GPA 3.86/4.0, Rank 4/187).

Research Interests

  • Trustworthy LLMs and reasoning robustness
  • Multimodal large language models (MLLMs)
  • Medical AI and clinically grounded multimodal reasoning
  • Controllable generation (TTS / animation)

Selected Publications

† equal contribution, * corresponding author

  1. Reasoning Hijacking: Subverting LLM Classification via Decision-Criteria Injection. ACL 2026.
  2. Human-MME: A Holistic Evaluation Benchmark for Human-Centric Multimodal Large Language Models. ICLR 2026.
  3. Segment-Aware Conditioning for Training-Free Intra-Utterance Emotion and Duration Control in Text-to-Speech. ACL 2026.
  4. Med-CMR: A Fine-Grained Benchmark Integrating Visual Evidence and Clinical Logic for Medical Complex Multimodal Reasoning. CVPR 2026.
  5. Soul: Breathe Life into Digital Human for High-fidelity Long-term Multimodal Animation. CVPR 2026.
  6. Multi-task learning for calcaneus fracture diagnosis of X-ray images. Biomedical Signal Processing and Control, 2025.

Experience

Machine Learning Engineer Intern, TikTok (ByteDance)

Feb 2026 – Aug 2026

  • Developed and deployed an automated compliance-review system with multimodal LLMs for Proof-of-Address verification.
  • Processed large-scale heterogeneous and adversarial real-world image data in security-sensitive production workflows.
  • Investigated interpretability and efficiency bottlenecks in black-box model deployment.

Golang Backend Developer, Shopee (Full-time)

Sep 2024 – Jun 2025

  • Designed and launched an automatic promotion-renewal system covering 5,000+ promotions and 20,000+ SKUs.
  • Reduced operation workload by approximately 50 person-hours per week.
  • Optimized discount-related partition strategy and indexes, reducing request latency to around one-tenth.

Research Intern, CAXA Technology

Jul 2023 – Nov 2023

  • Built a 60,000-scale 3D CAD surface-feature dataset through C++ plugin-based conversion.
  • Designed a GNN model for CAD face-feature recognition.
  • Supported generated wireframe fitting to 3D CAD models with pythonOCC.

Contact