Yuanhao Li — Research & Code

01A LITTLE ABOUT ME

Learning. Reasoning. Building.

I’m Yuanhao, a PhD student in Computer Science at the University of Exeter, supervised by Haozhe Wang and Geyong Min. I study how intelligent agents can learn, adapt, and make decisions in complex environments.

My research connects AI-native networks, reinforcement learning, and language model agents — from adaptive radio access network slicing to generalization in continuous control. Alongside research, I explore AI-assisted development and build tools through code.

learning, one experiment at a time.

Education

A path through data, computing, and intelligent systems.

2023 — 2027

University of Exeter

PhD · Computer Science

Expected September 2027 · UK

2022 — 2023

University of Edinburgh

MSc · High Performance Computing and Data Science

EPCC · UK

2018 — 2022

BNU–HKBU United International College

BSc · Data Science

Division of Science and Technology

Research interests

Three connected threads in my research.

Reinforcement learning

Multi-agent decision-making and structural generalization in continuous control.

AI-native networks

Learning-driven resource allocation and adaptive RAN slicing for next-generation networks.

Language model agents

Reflection, self-improvement, and language models for network control.

Publications & manuscripts

Published work and research in progress.

Published/01

A Flexible and Scalable Multi-Agent Learning Framework for Dynamic RAN Slicing in 6G Native-AI Networks

IEEE Transactions on Mobile Computing · Vol. 25, No. 5 · 2026

A scalable multi-agent reinforcement learning framework that adapts RAN slicing to changing demands without retraining, balancing service quality and reconfiguration overhead.

Read paper

Preprint · submitted to ICC 2027/02

Reflection-Driven Self-Improving Language Model Agents for Adaptive RAN Slicing

arXiv

An Actor–Reflector framework that combines online interaction, semantic reflection, and preference learning to improve network control without manually designed reward functions.

Read paper

Under review · ICLR 2027/03

CrossMaze: Benchmarking Zero-Shot Structural Generalization in Offline Continuous Control

OpenReview

A benchmark for generalization to unseen maze layouts, with a compact language-model-based policy that maps layouts, states, and goals to continuous actions.

Read paper

Technical toolkit

Tools I work with, from experiments to implementation.

Programming & ML

Python
R
PyTorch
Transformers
C++
ns-3
Linux
Docker

High-performance computing

MPI
OpenMP
CUDA
Slurm
DeepSpeed
Multithreading / Multiprocessing

AI-assisted development

Claude Code
Pi
Codex

Honors & awards

China Scholarship Council (CSC) scholarship for doctoral study abroad
Second-Class Scholarship during undergraduate studies

FIND ME ELSEWHERE

Let’s connect.

For conversations about research, intelligent agents, or building things.

yl1118@exeter.ac.uk

Alternative emails

Learning. Reasoning. Building.

Education

University of Exeter

University of Edinburgh

BNU–HKBU United International College

Research interests

Reinforcement learning

AI-native networks

Language model agents

Publications & manuscripts

A Flexible and Scalable Multi-Agent Learning Framework for Dynamic RAN Slicing in 6G Native-AI Networks

Reflection-Driven Self-Improving Language Model Agents for Adaptive RAN Slicing

CrossMaze: Benchmarking Zero-Shot Structural Generalization in Offline Continuous Control

Technical toolkit

❯Programming & ML

❯High-performance computing

❯AI-assisted development

Honors & awards

Let’s connect.

Programming & ML

High-performance computing

AI-assisted development