Lu Liu at an exhibition with the Department of Health, on behalf of the Centre for Artificial Intelligence (now the Australian Artificial Intelligence Institute).
What is the most rewarding aspect of your research?
Learning to learn is a cutting-edge machine learning technology that is well-recognised as one of the pathways towards Artificial General Intelligence. In particular, it aims to develop a meta learning module that can rapidly adapt deep learning models using very few samples. My research is to augment the meta learning process by exploiting the structured relationship among objects (i.e. classes in classification task). As evaluated in the image recognition task, my proposed method significantly boosts the performance of meta learning in the context of few-shot learning scenario. My research work has been published in a few top-tier conferences and journals including NeurIPS, AAAI, IJCAI, and IEEE TKDE. More details can be found at my personal homepage.
What are the real-world applications of your research?
My research focus on the fundamental topic of machine learning, and my research outcomes can be applied to a broad range of applications. Below are some examples for how the meta learning could be applied to real-world applications.
- Edge device computing and personalization: Personal devices suffer from the challenges of few-data. It is interesting to see how to quickly adapt a deep learning model to users and provide personalized services.
- Data privacy protection: Without uploading users’ data to the server, few-shot learning can provide services trained on deep learning models without the sacrifice of user privacy.
- Cold-start recommendation: Analyse behaviours of new users with few data available is important towards user retention.
What ideas do you have for future research?
We would like to expand my research from both theoretical and empirical ways. Firstly, we will broaden the idea of meta-learning or learning to learn on broader range of challenging scenarios. Secondly, we will explore more collaborations with industry to customise my methods to real-world applications.
Are you involved in collaborative research?
Within Australia, I have been involved with Professor Fethi Rabhi and Professor Lina Yao from University of New South Wales. Internationally, we are collaborating with Professor William Hamilton from University of McGill, Canada, Hugo Larochelle from Google Brain, Canada and Tianyi Zhou from University of Washington, USA.
What inspired you to undertake a PhD in computer science?
Computer science is a promising field, and machine learning (ML) has been shining its light for the past decades. I believe the era of ML is not ending yet, and will remain as one of the most important factors to influence every part of our society today.