Qingyu Qu (屈清宇)
Hi! I'm Qingyu Qu (屈清宇), a researcher interested in the learning and representation of natural data, especially by artificial and biological neural networks. I am currently pursuing my PhD in the Liu Ziming AI Science Research Group and also work at MetaCircle.
Follow my updates on GitHub.
Education
- Undergraduate, Physics, Southern University of Science and Technology (SUSTech)
- Intern, Chinese Institute for Brain Research (CIBR)
- Visiting Student, Physics, UC Berkeley
- Pursuing PhD, College of AI, Tsinghua University
Publications
Scaling of Relaxation and Entropy in Buckled Colloidal Monolayers
Physical Review Letters 137, 048201, 2026 (Editors' Suggestion)
PDF
Experiments
- Quantifying Structure in Data — Piecewise-linear estimation of the structural fraction in toy setups (in Chinese).
- Quantifying Structure in Data — Effect of Model Width — Two-cluster teacher-student protocol with teacher/student width swept over {4,…,128}: C~N curves per width and post-knee slope jitter vs width (in Chinese).
- Induction Circuit — Causal verification of an induction circuit in a two-layer toy transformer: head scores, global ablations, vector-injection attention maps, and a minimal three-head induction machine on synthetic and real text (in Chinese).