Haodong Xie

I am currently a Research Fellow at the University of Warwick, working on the KnoWare project, supervised by Professor Thomas Hills. My research focuses on LLM-based agents, emergent capabilities, self-improvement, and metacognition, with a broader interest in understanding the behaviour, generalisation, and robustness of foundation models.

Previously, I was a PhD student in Computer Science at the University of Manchester, supervised by Professor Angelo Cangelosi. I also hold an MSc in Advanced Computing from King’s College London and a BEng in Internet of Things from Hohai University in China.

Email / Google Scholar / Linkedin

Portrait of Haodong Xie

News

- HIL-CBM got accepted by BMVC 2026.

- Start my role as a Research Fellow at the University of Warwick.

- Presented two posters at CogSci 2026 in Rio de Janeiro.

- PhD in Computer Science Awarded by the University of Manchester.

Selected Publications

A full publication list is available on Google Scholar.

Architecture diagram of the Hierarchical, Interpretable, Label-Free Concept Bottleneck Model

Hierarchical, Interpretable, Label-Free Concept Bottleneck Model

Haodong Xie, Yujun Cai, Yiwei Wang, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi

British Machine Vision Conference, 2026.

[paper]

Linear-probing accuracy across layer depth for several LLMs distinguishing concreteness

Distinguishing Concreteness Differences in LLM Representations via Linear Probing

Haodong Xie, Yuze Sha, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi

Annual Meeting of the Cognitive Science Society, 2026.

[paper]

Concept learning across superordinate, basic, and subordinate levels of abstraction

Learning Abstract Categories through Linguistic Feature Explanations

Haodong Xie, Xuena Wang, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi

Poster with abstract, Annual Meeting of the Cognitive Science Society, 2026.

Multimodal generative approach to abstract concept learning across naming and understanding

From Concrete to Abstract: A Multimodal Generative Approach to Abstract Concept Learning

Haodong Xie, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi

International Conference on Neural Information Processing, 2024.

[paper]

Projects & Outreach