Undergraduate researcher at the University of Waterloo, working with Prof. Wenhu Chen at TIGER-AI-Lab.
🌐 Website: mingyisu.github.io/mingyisu · 🎓 Google Scholar · 🤗 Hugging Face · 💼 LinkedIn · 📫 m7su@uwaterloo.ca
- Trustworthy AI agents: self-verification, error recognition, and reliable task completion
- Reward modeling & RL for more reliable agent behavior
- Evaluation & benchmarking of LLMs and AI agents
- Multimodal representation learning & retrieval across text, image, video, and speech
- MMEB-V3: Measuring the Performance Gaps of Omni-Modality Embedding Models. COLM 2026 · Paper
- VLM2Vec-V2: Advancing Multimodal Embedding for Videos, Images, and Visual Documents. TMLR 2026 · Paper · Code · Leaderboard
- BrowseComp-Plus: A Fair and Disentangled Evaluation Benchmark for Deep Search Agents. ACL 2026 · Paper
- B3: Breaking the Batch Barrier of Contrastive Learning via Smart Batch Mining. NeurIPS 2025 · Paper
- VideoScore2: Think before You Score in Generative Video Evaluation. arXiv 2025 · Paper
- Trustworthy Medical AI: Multi-Judge Evaluation & Alignment: multi-generator, multi-judge LLM evaluation for medical documentation, with preference-based fine-tuning of Qwen3-4B
- ECG Digitization from Photographs: extracts ECG waveforms from photos using a CNN with spatial attention
Python · PyTorch · Transformers · TRL · LangChain · R · C++ · SQL
💬 I'm looking for graduate research opportunities. Feel free to reach out!