Hi! I am Feifei Qian (Sophia), a Ph.D. student in the School of Artificial Intelligence at Beijing Normal University, advised by Prof. Lu Bai.
My research interests span graph machine learning, LLM evaluation, and LLM post-training, with a particular focus on:
- Graph Machine Learning: Studying graph kernels and the expressive power of graph neural networks, with an emphasis on structural encoding, information preservation, and over-smoothing.
- LLM Rubrics: Understanding how rubric design shapes evaluation and learning, developing benchmarks for systematic analysis, and investigating the role of rubric-based rewards in LLM post-training.
🔥 News
- 2026.09: 🎉 Our paper, “Beyond Coordinates: Encoding Graph Structure via Contextual Distribution and Relational Similarity,” has been accepted to NeurIPS 2026 as a Spotlight!
- 2026.05: Joined the Language and Intelligence Group at Beike as an LLM Algorithm Intern, working on agent training and evaluation.
- 2025.05: One paper has been accepted to IJCAI 2025!
- 2024.12: One paper has been accepted to AAAI 2025!
📝 Selected Publications
From Structure to Validity: Mechanism-Guided Generation and Refinement of LLM Rubrics
Feifei Qian, Xuan Chen, Lixin Cui, Chaoyang Mei, Fanlin Meng, Shuyao Xiao, Ming Cui, Junxi Yin, Lu Bai
- ARTS-Hybrid combines adaptive rubric tree construction with semantic-first, response-calibrated refinement to improve rubric quality and LLM evaluation.
Beyond Coordinates: Encoding Graph Structure via Contextual Distribution and Relational Similarity
Feifei Qian, Lu Bai, Lixin Cui, Ming Li, Hangyuan Du, Edwin Hancock
- EntSE and KerSE encode graph structure through contextual spectral distributions and relational similarity, connecting local information with expressive graph representations.
Feifei Qian, Lu Bai, Lixin Cui, Ming Li, Hangyuan Du, Yue Wang, Edwin Hancock
- An entropy-based perspective on graph-level over-smoothing, with sampling and feature discretization to preserve information in high-entropy regions.
Feifei Qian, Lu Bai, Lixin Cui, Ming Li, Ziyu Lyu, Hangyuan Du, Edwin Hancock
- Hierarchical substructure composition with entropy regularization, feature-level attention, and kernel-level attention for end-to-end graph classification.
🎖 Honors and Awards
- 2022: National Scholarship (Top 1%).
- 2021: Huawei Scholarship.
🤝 Academic Service
Reviewer
- Journals: Pattern Recognition; IEEE Transactions on Neural Networks and Learning Systems (TNNLS); Neural Networks.
- Conferences: NeurIPS 2026 (Top Reviewer); ICLR 2027; WWW 2024.
📖 Education
- September 2023 – June 2028 (expected), Beijing Normal University. Ph.D. student
- September 2019 – June 2023, Anhui University. Bachelor’s degree in Computer Science and Technology.
💻 Internships
- May 2026 – October 2026, Beike, Language and Intelligence Group, LLM Algorithm Intern.
- Developed agent data pipelines and two-stage supervised fine-tuning.
- Built rubric-driven evaluation for AIGC agent workflows, covering tool-use trajectories, execution constraints, and delivery quality.
- October 2022 – May 2023, NIO, Software Engineering Intern.
- Contributed to engineering tooling and quality assurance workflows, supporting software reliability and development efficiency.
📚 Technical Blog
- Do Rubrics Expand What RL Can Learn? — Exploring how rubrics influence reinforcement learning and reasoning capabilities in LLM post-training.
- Browse the technical notebook →