ESROP + A research-oriented project to study diffusion-based LLMs

This project studies a promising alternative to current autoregressive (AR) large language models: diffusion-based large language models. It aims to make them generate high-quality text more efficiently while maintaining speed and energy efficiency. The research will focus on understanding how these models work, identifying their root limitations, and developing improved methods that balance text quality, speed, and energy use. Through a combination of literature review, theoretical analysis, and experiments, the project will propose and evaluate new approach for faster and more reliable text generation. The participating institutions will benefit by advancing their expertise in AI models, strengthening international research collaboration, and producing research outcomes that can support future work in efficient and scalable language models.

Faculty Supervisor:

Arthur Chan

Student:

Partner:

King Mongkut’s University of Technology Thonburi

Discipline:

Engineering

Sector:

Artificial Intelligence; Technology; Information and Communications Technology (ICT)

University:

University of Toronto

Program: