Liang Zhao

Current work and legacy

• Serves as a Senior Lecturer/Associate Professor at UTS, mentoring next-generation AI researchers.
• Advocates for ethical AI development, emphasizing fairness and societal impact.
• His work bridges theoretical AI and real-world applications, influencing industries like healthcare, environmental science, and smart cities.

Generative AI on complex data

Zhao has conducted research on complex data deep modeling, focusing on spatial, temporal, networked, textual, and heterogeneous types. He advanced new graph neural networks and inference strategies to encode complex data into their components, enabling independent and combined learning of their embeddings.^[14] To contribute to the deep graph learning research domain, he published Graph Neural Networks: Foundations, Frontiers, and Applications with Jian Pei, Peng Cui, and Lingfei Wu which covered a range of topics in deep learning on graphs.^[15]

Zhao jointly characterized the distributions of temporal and network aspects using new techniques in temporal random walk generation and end-to-end walk assembly.^[16] His contributions include creating deep generative models for spatial networks that identify the interplay between spatial and network factors, as well as factors related solely to spatial or network information.^[17] Further enhancing deep graph transformation, which generated target graphs conditioned on source graphs, he used applications such as molecule structure optimization and circuit obfuscation with his work on deep generative models for graphs extending into several directions, including property-controllable complex data generation and design.^[18] While addressing the critical need for unique datasets and model evaluation strategies in deep generative models, he released benchmark dataset repositories such as GraphGT at NeurIPS 2021,^[19] along with review papers on the method categorization and standardization.^[20]

Collaborative machine learning strategies

Zhao focused on developing models for learning and predicting across both known and unknown tasks. His research introduced directions in spatial multi-task learning, balancing the trade-offs between task relations and differences, as well as spatial correlation and heterogeneity.^[21] Beyond traditional approaches, he extended multi-task learning frameworks into domain generalization scenarios, allowing models trained on known tasks to generalize to unseen tasks across different locations and times.^[22][23] He then conducted research on continual learning integrated temporal considerations to manage the persistence and adaptability of a model's memory across old and new tasks, drawing insights from neuroscience principles of memory retention to achieve high accuracy with low computational costs.^[24]

Human-AI interaction and alignment

Zhao investigated explainable AI, particularly AI reasoning and its correction through human guidance, to improve medical imaging for disease diagnosis by developing techniques, benchmarks, and evaluation scenarios in collaboration with radiologists and clinicians. He focused on proposing methods that narrow the distributional gaps between humans and AI's explanations in complex data such as 2D/3D images, graphs, and spatiotemporal data^[25][26][27] which can benefit both training and promotion of AI models.^[28] Additionally, his team developed and deployed user interface systems that enables the online interaction between human and AI.^[29]

Key publications

• Authored influential papers in top-tier journals/conferences (e.g., IEEE Transactions on Neural Networks, KDD, AAAI).
• Co-authored Deep Learning for Spatio-Temporal Data Mining (2021), a seminal text on integrating deep learning with spatial-temporal analytics.

Selected books

• Graph Neural Networks: Foundations, Frontiers, and Applications (2022) ISBN 978-981-16-6056-6

Selected articles

• Zhao, L. (2021). Event prediction in the big data era: A systematic survey. ACM Computing Surveys (CSUR), 54(5), 1-37.
• Ling, C., Zhao, X., Lu, J., Deng, C., Zheng, C., Wang, J., ... & Zhao, L. (2023). Domain specialization as the key to make large language models disruptive: A comprehensive survey. arXiv preprint arXiv:2305.18703.
• Zhao, L., Sun, Q., Ye, J., Chen, F., Lu, C. T., & Ramakrishnan, N. (2015). Multi-task learning for spatio-temporal event forecasting. In Proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining (pp. 1503–1512).
• Chai, Z., Chen, Y., Anwar, A., Zhao, L., Cheng, Y., & Rangwala, H. (2021). FedAT: A high-performance and communication-efficient federated learning system with asynchronous tiers. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (pp. 1–16).
• Ling, C., Jiang, J., Wang, J., Thai, M. T., Xue, R., Song, J., ... & Zhao, L. (2023). Deep graph representation learning and optimization for influence maximization. In International Conference on Machine Learning (pp. 21350–21361). PMLR.