Learning complex dynamics from multimodal time-series data

Overview

Modelling non-stationary dynamics from high-frequency time-series data remains challenging. These signals often exhibit complex temporal and spectral structure, while observations are typically noisy, incomplete, and affected by changing operating conditions, making reliable prediction and representation learning difficult.

This PhD project, offered at Queensland University of Technology (QUT) in collaboration with industry partners, focuses on learning representations and dynamics from multimodal time-series data.

The research will explore deep approaches including sequence models, transformer-based architectures, anomaly detection, graph neural networks, and self-supervised learning, with an emphasis on generalisation, robustness, and data efficiency.

Applications include real-world sensing systems, such as large-scale IoT networks, and power quality monitoring.

Skills and experience

• Strong background in programming (preferably Python).
• Experience in developing and training deep machine learning models (e.g. PyTorch).
• Interest in time-series modelling, representation learning, and self-supervised learning.
• Strong problem-solving skills and the ability to work independently and collaboratively.
• Background in electrical engineering/civil engineering/computer science, or a related field, with a focus on machine learning or data-driven methods.

Scholarships

You may be eligible to apply for a research scholarship.

Explore our research scholarships

Keywords

• deep learning
• self-supervised learning
• time-series data
• multimodal learning
• anomaly detection

Contact

Contact the supervisor via email maryam.haghighat@qut.edu.au for more information.