SIRCL: Stabilizing Implicit Reasoning via Centripetal Latent Trajectories
A lightweight geometric regularization method for stabilizing implicit reasoning trajectories without adding inference-time overhead.
🎓MPhil Student in Artificial Intelligence
Hong Kong University of Science and Technology (Guangzhou)
DIGAI Lab
MPhil in Artificial Intelligence
2025-08-01
2027-07-31
Hong Kong University of Science and Technology (Guangzhou)
BEng Computer Science and Technology
2020-09-01
2024-06-30
Shandong University
I work on implicit chain-of-thought reasoning, continuous latent-space reasoning, and language model reasoning systems.
My current projects study how latent tokens behave during multi-step reasoning, how to stabilize continuous reasoning trajectories, and how to improve the representational capacity of implicit reasoning.
This site collects my publications, research projects, technical notes, and selected writing.
A lightweight geometric regularization method for stabilizing implicit reasoning trajectories without adding inference-time overhead.
A spectral regularization method for improving latent-token diversity and mitigating rank collapse in implicit reasoning.
Publish your data science and research directly from Jupyter Notebooks. No screenshots required.
Take full control of your personal brand and privacy by migrating away from the big tech platforms!
Create a personal knowledge base and share your knowledge with your peers.
Use popular tools such as HuggingFace, Plotly, Mermaid, and data frames.
Embed videos, podcasts, code, LaTeX math, and even test students!
Easily manage your projects - create ideation mind maps, Gantt charts, todo lists, and more!