Yupei Li

Yupei Li

PhD Student in Computing @ Imperial College London

✉️ yupei.li22@imperial.ac.uk 🎓 Google Scholar 🔗 LinkedIn 🌐 Github 📍 London, UK

Bio

I am a second-year PhD student at Imperial College London, supervised by Prof. Björn Schuller and Prof. Lucia Specia. My research focuses on multimodal AI-generated content detection, with particular interest in the intersection of emotion recognition, LLMs reasoning and human-inspired neural network research.

Prior to my PhD, I obtained my MRes and MSc in Computing (AI & ML) from Imperial College London, and a B.S. in Computer Science (Honours) from Xi'an Jiaotong University (Qian Xuesen Pilot Class, top 5%). I was also a visiting student at the University of Oxford through Lady Margaret Hall's exchange programme. I have also the main proposal writer to support my Supervisor Björn Schuller to acquire TUM Global Incentive Fund in 2025.

News

Education

Imperial College London, UK 10/2024 – 10/2027
PhD in Computing
Supervisor: Prof. Björn Schuller & Prof. Lucia Specia
Thesis: Identification of Multimodal AI-generated Content
Imperial College London, UK 10/2023 – 10/2024
MRes in Computing (AI & ML) — Distinction
Supervisor: Prof. Lucia Specia
Thesis: Discourse Feature Enhanced AI-generated Text Detection
Imperial College London, UK 10/2022 – 10/2023
MSc in Computing (AI & ML)
Thesis: Translation Tasks with Multi-lingual Corpus and Non-autoregressive Model
ISO: CNLP (Computer Vision and NLP) in Sentiment Analysis
University of Oxford, UK 10/2021 – 06/2022
Visiting Student in Computer Science
Lady Margaret Hall Exchange Programme
Xi'an Jiaotong University, China (C9 League) 09/2019 – 07/2022
B.S. in Computer Science (Honours)
Qian Xuesen Pilot Class (40 students selected per year)
GPA: 89.2/100 (top 5%) · Graduated one year in advance

Research & Publications

🔎 AI-Generated Audio & Text Detection
Emotion-Informed Training Framework for Audio Deepfake Detection
Yupei Li, et al.
Under review at Interspeech 2026
PDF Code
Generalized Audio-LLM Framework for Robust Deepfake Detection and Spoof Attribution
Yupei Li, et al.
Under review at Speech Processing Letters
PDF Code
Discourse Relations and Human-Written Style for Detecting AI-Generated Text
Yupei Li, et al.
Accepted at IJCNN 2025
PDF Code
🧠 Transfer Learning & LLM Post-Training
Structural Alignment Between Code and Math for Autoformalisation
Yupei Li, et al.
Under review at EMNLP 2026
PDF Code
🔬 Human-inspired neural network research
Brain-Inspired Neural Network Learning: Neuroplasticity, Neurotransmitters with Dropin and Dropout
Yupei Li, et al.
Paper accepted at IJCNN 2026
PDF

Internships

Tencent Co., Ltd 05/2026 – Present
Machine Learning Engineering · Mentor: Zhaopeng Tu
Building up digital human systems with Hunyuan LLMs.
LLMs SFT DPO
CUHK (Chinese University of Hong Kong) 06/2025 – 11/2025
Research Assistant · Mentor: Zhizheng Wu
Audio deepfake detection with LLM. Analyzed the impact of audio encoders and text-based LLMs in Audio-LLM architectures. Proposed a generalized ALLM framework achieving state-of-the-art performance across multiple datasets.
LLMs Audio Deepfake Detection
Alibaba Co., Ltd 08/2025 – 12/2025
NLP Research Intern · Mentor: Chenyang Lv
Emotion recognition and deepfake detection on audio. Developed an emotion-informed training framework bridging conventional deepfake features with emotion representations, achieving 6% and 2% accuracy increases on two benchmarks.
Audio Processing Emotion Recognition Deepfake Detection
Huawei London Co., Ltd 01/2025 – 07/2025
NLP Research Intern · Mentor: Philipp Borchert
Transfer learning from code to math autoformalisation. Discovered structural alignment between code and math, created a novel multi-stage training approach. Achieved BEQ and typecheck improvements over simple math dataset training.
LLMs Transfer Learning Math Reasoning
Shenzhen Futu Network Technology Co., Ltd 07/2022 – 09/2022
NLP Algorithm Engineer
Designed query strategies for NER problems. Compared three methods of entropy calculation: BERT decoder, cosine similarity, and LLAL with Global Pointers. Built a web pipeline combining active learning and NER model.
NLP NER Active Learning Financial Knowledge