Advancing Reasoning Ability in Diffusion Large Language Models

This PhD project focuses on advancing diffusion large language models with reasoning ability, aiming to leverage the rich semantic understanding of language models to provide human-like reasoning ability across various domains. The research will explore how language models, with their ability to process complex language inputs and reason the output, can be effectively obtained by training on different tasks with a wide range of datasets. Key challenges include enhancing scalability for real-time reasoning, improving model interpretability, and ensuring robustness and faithfulness in reasoning generations. The project will also investigate novel ways to utilise math, coding, or legal data and reinforcement learning fine-tuning strategies to tailor language models for advanced reasoning.

The Data Science (DS) Discipline at EECS has been largely responsible for UQ's rating of 5 (well above the world standard) in the last 3 consecutive Australia ERA exercises, for the field of Information Systems. It has a strong and internationally focused research culture as well as a growing portfolio of datasets and data-intensive computing infrastructure capable of supporting terabyte-level high-performance storage and processing. DS members have advised ~100 RHD students in the last 5 years and contribute significantly to the national and international research community in various leadership roles. PhD students in DS will be well-supported with full access to the advanced computing facilities and the international collaboration network of the supervisors.

This project is supported by the Research project scholarship.

Learn more about the Research project scholarship.

Your application will be assessed on a competitive basis.

We take into account your:

• previous academic record
• publication record
• honours and awards
• employment history

A working knowledge of natural language understanding and reinforcement learning would be of benefit to someone working on this project.

You will demonstrate academic achievement in the field/s of natural language understanding and RL, and the potential for scholastic success.

A background or knowledge of natural language understanding and reinforcement learning is highly desirable.

You must submit an expression of interest (EOI) by 27 October, 2026 27 October, 2026.

Before you apply

1. Check your eligibility for the Doctor of Philosophy (PhD).
2. Prepare your documentation.
3. If you have any questions about whether the project is suitable for your research interests, contact Dr Ruihong Qiu (r.qiu@uq.edu.au).

When you apply

To apply, submit an expression of interest (EOI) for the program. You don't need to apply separately for the project or scholarship. How to submit an EOI

In your EOI, complete the ‘Scholarship/Sponsorship’ section with the following details:

1. Are you applying for an advertised project: 'Yes'
2. Project: 'Research project scholarship'
3. Scholarship Code Listed in the Advertisement: DIFFUSION-QIU
4. Link to Scholarship Advertisement: https://study.uq.edu.au/study-options/phd-mphil-professional-doctorate/projects/advancing-reasoning-ability-diffusion-large-language-models

Submit an EOI