Dr Weijia Zhang

Dr Weijia Zhang’s research aims to bridge the gap between causal inference and machine learning, which enables statistical and computational algorithms to transform the decision-making process of our government, healthcare, and business.

Dr. Weijia "Weja" Zhang graduated with a PhD degree from the University of South Australia in 2018. He has researched and taught a variety of courses in data science and machine learning, first when working as a Research Fellow in his alma mater, and then as an Associate Professor at Southeast University, China, before assuming a position as Lecturer with the University of Newcastle.

As a Lecturer in Data Science and Statistics at the University of Newcastle, Weijia is experienced in statistical causal inference and weakly supervised machine learning. His research has been featured in many prestigious international conferences and journals, and has been successfully adapted to various industries, including some major companies in the automobile, oil and gas, and telecommunications sectors in the Asia Pacific region.

The power of causal inference.

Causal inference is a field of study that deals with the identification of cause-and-effect relationships between variables. It has had a significant impact in various fields, including medicine, public health, economics, and social sciences, among others. Weijia’s work in causality focuses on developing methods to estimate the causal effect from observational data or even data with censoring, which allows researchers to determine the relationship between an exposure or intervention and an outcome of interest, without the need for running costly and time-consuming randomized controlled trials.

Machine learning from ambiguously labelled data

Although supervised machine learning has demonstrated success in various domains, it has a notable limitation in its dependence on large volumes of labelled data. Weakly supervised learning, which involves training algorithms using data with incomplete, ambiguous or noisy labels, is a technique employed in machine learning to overcome this limitation. Weijia's research in weakly supervised learning has sparked a new learning paradigm, known as Multi-Instance Partial Label learning, which enables algorithms to learn from low-quality and ambiguous labels provided by non-expert crowdsourced annotators. This paradigm offers a promising solution to the challenge of limited expert annotations in crucial domains, such as medical imaging and diagnostics.

Bridging causal inference with machine learning

It comes as no surprise that Weijia's research in causal inference and machine learning leads him inevitably to the emerging field of causal machine learning. This area has the potential to enhance the dependability and interpretability of machine learning models, which have impeded the advancement of machine learning applications since its inception. Thanks to Weijia's expertise in weakly-supervised learning, one of his recent works has demonstrated that causality-based models can be more readily achieved when supervision is ambiguous than in traditional supervised learning.

Making realworld impact

In addition to his academic publications, Weijia has extensive experience in translating his research into practical applications. He has collaborated with customer representatives in the telecommunications sector to enhance customer retention, and with production engineers in the oil and gas industry to develop predictive maintenance strategies, among other projects. In a recent collaboration with Chang'an Automobile, a major car manufacturer in the Asia Pacific, Weijia's algorithm was implemented to improve the lifespan for hundreds of thousands of vehicle batteries.