👋 Welcome to my website!

My name is Yifan Li (李一帆). I am currently a senior undergraduate student at Qinghai University.

I will soon join the School of Computer Science and Technology, Beijing Institute of Technology (BIT) as a Ph.D. student, under the supervision of Prof. Ying Fu (付莹) and Prof. Junliang Xing (兴军亮).

🔬   My previous research mainly focused on audio separation and multimodal retrieval. Currently, I am exploring interesting research problems and new directions, and I will keep updating this website with my latest research and progress.

🤗   I will open-source my research projects and code whenever possible.

📮   Feel free to contact me for research discussions and collaborations.

🔥 News

• 2026.09: 🎉 Our paper Falcon was accepted by NeurIPS 2026!
• 2026.04: 🥈 We won the Second Prize in the China National Finals of the 10th Huawei ICT Competition!
• 2026.03: 🎉 My first-author paper MPS-CLIP was accepted by ICME 2026!
• 2026.03: 🥇 We won the First Prize in the ASC 2026 Student Supercomputer Challenge!
• 2026.01: 🚀 My research project on multi-instrument music source separation was successfully funded by Qinghai University.
• 2025.02: 🥈 We won the Second Prize in the ASC 2025 Student Supercomputer Challenge!
• 2025.01: 🚀 My research project on LLM-assisted remote sensing image understanding was successfully funded as a key research project at Qinghai University.

📝 Publications

NeurIPS 2026

Efficient Streaming Audio-Visual Target Speaker Extraction for Real-World Acoustic Scenes

Wendi Sang^*, Kai Li^*, Yifan Li, Jianqiang Huang, Xiaolin Hu. Conference on Neural Information Processing Systems (NeurIPS) 2026. Sydney, Australia.

• Falcon enables streaming audio-visual target speaker extraction with a single-pass multi-scale separator and speaker-contrastive training. We introduce RealSSA, combining realistic simulated scenes with real-recorded evaluation. Compared with causal AV-TFGridNet, Falcon uses 91.9% fewer parameters, 20.2× less computation, and 11.1× lower GPU latency.

| | | |

ICME 2026

Multi-Perspective Subimage CLIP with Keyword Guidance for Remote Sensing Image-Text Retrieval

Yifan Li^*, Shiying Wang^*, Jianqiang Huang. IEEE International Conference on Multimedia and Expo (ICME) 2026. Bangkok, Thailand.

• MPS-CLIP improves remote sensing image-text retrieval through keyword-guided multi-perspective alignment. G²A and MPR combine global context and local cues while keeping the backbone frozen. It achieves 35.18% mR on RSICD and 48.40% mR on RSITMD.

| | |

🎖 Honors and Awards

• 2026.04 🥈 Second Prize, 10th Huawei ICT Competition, China Final
• 2026.03 🥇 First Prize, ASC Student Supercomputer Challenge
• 2025.02 🥈 Second Prize, ASC Student Supercomputer Challenge

📖 Educations

• 2023.09 - now, Undergraduate, Department of Computer Technology and Application, Qinghai University, Xining.