Scientific Manuscript Audit
A structured audit workflow for Codex and Claude Code that traces scientific claims to evidence and tests what matters for submission and revision.
Machine Learning · Stochastic Dynamics · Theoretical Physics
Kangqiao Liu 刘康桥
I study how learning and physical systems move through noisy, high-dimensional landscapes. My work spans stochastic optimization, nonequilibrium physics, quantum information, and complex dynamics, with current interests in reasoning and AI-assisted scientific discovery.
Academic profiles
Research landscape
Dynamics across learning and physical systems.
The subjects have changed over time, but the questions have stayed surprisingly similar: what sets fluctuations, response, escape, transport, and instability?
01 / LEARNING Stochastic learning and optimization Finite-learning-rate SGD, minibatch noise, escape dynamics, and the stochastic structure of learning. 02 / NONEQUILIBRIUM & QUANTUM Information, response, and transport Kinetic uncertainty relations, quantum response bounds, information engines, random access codes, and constrained transport. 03 / COMPLEX DYNAMICS Instability and chaos Lyapunov bounds, competing dynamical scales, and instability in gravitating systems.
Selected work
A few points along the research trajectory.
These papers show the thread from learning dynamics to nonequilibrium response and quantum transport. The publications page contains the complete record.
2021 · MACHINE LEARNING Noise and Fluctuation of Finite Learning Rate Stochastic Gradient Descent 2025 · NONEQUILIBRIUM DYNAMICS Dynamical activity universally bounds precision of response in Markovian nonequilibrium systems 2026 · QUANTUM DYNAMICS Maximal-velocity deficit under a finite-support constraint in a hard-wall half-line continuous-time quantum walk
Research tooling
Tools for examining and testing ideas.
Research workflows and interactive scientific resources, with methods and examples available for inspection and independent study.
Advanced Quantum Mechanics
Nine chapters of graduate-level lecture notes and 35 browser-based physics investigations, pairing derivations with reproducible numerical checks and questions for independent, AI-assisted study.
Affiliation
Xihua University
Key Laboratory of High Performance Scientific Computation
物理学讲师 · 西华大学理学院
Contact
Email
kqliu-AT-xhu.edu.cn (replace -AT- by @)
Office
6D416
Address
Chengdu, Sichuan, China 610039