Thesis Title
Large Language Models (LLMs) for Multi-variate Time-series (MVTS) Generation

Research Description
Du’s research focuses on developing a multi-modal large language model (MMLLM) that effectively interacts with domain-specific multi-variate time-series (MVTS) in fields like traffic, electricity, and smart buildings, etc. The goal is to create a model capable of understanding and generating MVTS, facilitating intuitive interaction for both experts and laypeople. Key tasks include MVTS-based Q&A, generating MVTS from textual inputs, and advanced capabilities like large-scale MVTS generation and integrating additional data modalities such as weather. This research aims to pioneer new simulation-in-the-loop paradigms for urban scenarios, enhancing spatio-temporal model applications.

Supervisors
Prof Flora Salim, University of New South Wales