Lingqiao Liu

Lingqiao Liu is a computer vision and machine learning researcher at Adelaide University whose work runs from Fisher vector image coding to vision-language models and robot manipulation. He earned a PhD at the Australian National University in 2014 and received an ARC Discovery Early Career Researcher Award in 2016. His papers include a Fisher vector coding study in IEEE TPAMI in 2017, R2GenGPT for radiology reports in 2023, the continual-learning method CoLeCLIP in 2024, Chain-of-Action for robot manipulation at NeurIPS 2025, and a graph anomaly detection paper in IEEE TKDE in 2026. He is an Associate Professor in the School of Computer Science and Information Technology at Adelaide University and an academic member of the Australian Institute for Machine Learning.

Now at: Adelaide University and Australian Institute for Machine Learning

Timeline

2014–2016

ANU PhD and a DECRA

He earns a PhD from the Australian National University in 2014. In 2016, he receives an ARC Discovery Early Career Researcher Award (DECRA) and a University of Adelaide Research Fellowship, and his DECRA project is Zero-shot and few-shot learning with deep knowledge transfer.
2017

Fisher vector coding in TPAMI

In 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence publishes Compositional model based Fisher vector coding for image classification, which he coauthors with Peng Wang, Chunhua Shen, Lei Wang, Anton van den Hengel, Chao Wang and Heng Tao Shen. The method drops the usual assumption that a local feature comes from one of a few Gaussians and models its mean as a linear combination of key components.
2023

R2GenGPT for radiology reports

In 2023, he coauthors R2GenGPT with Zhanyu Wang, Lei Wang and Luping Zhou, a method that aligns visual features with the word embedding space of large language models to generate radiology reports. The language model stays frozen while a lightweight visual alignment module is trained.
2023–2025

Compositional zero-shot learning

He coauthors Learning Conditional Attributes for Compositional Zero-Shot Learning, accepted at CVPR 2023, which learns attribute embeddings conditioned on the recognized object, since an attribute such as "wet" differs between "wet apple" and "wet cat". In 2025, IEEE Transactions on Pattern Analysis and Machine Intelligence publishes his coauthored paper Learning Dual-Stream Conditional Concepts in Compositional Zero-Shot Learning.
2024–2025

CoLeCLIP for continual learning

In 2024, he coauthors CoLeCLIP with Yukun Li, Guansong Pang, Peng Wang and colleagues, a method for open-domain continual learning of vision-language models built on CLIP. It jointly learns a set of task prompts and a cross-domain class vocabulary and outperforms state-of-the-art methods on 11 domain datasets. IEEE Transactions on Neural Networks and Learning Systems publishes it in 2025.
2025

Chain-of-Action at NeurIPS

In 2025, he coauthors Chain-of-Action with Wenbo Zhang, Xiao Ma and others, a robot manipulation policy that generates a whole trajectory by reasoning backward from a task goal. It appears in the NeurIPS 2025 main conference and outperforms imitation learning methods such as ACT and Diffusion Policy across 60 RLBench tasks and 8 real-world tasks.
2026

Graph anomaly detection in TKDE

In 2026, IEEE Transactions on Knowledge and Data Engineering publishes his coauthored paper Learning Feature Encoder With Synthetic Anomalies for Weakly Supervised Graph Anomaly Detection.
As of October 2026

Associate Professor at Adelaide University

He is an Associate Professor in the School of Computer Science and Information Technology at Adelaide University and an academic member of the Australian Institute for Machine Learning. His current work covers computer vision, large language models and agents.