Xinyue XU

I am currently a PhD student at The Hong Kong University of Science and Technology, working under the supervision of Prof. Xiaomeng Li. Prior to this, I earned my Bachelor degree (with Honours) from The Australian National University, where I was fortunate to be advised by Senior Prof. Amanda Barnard and supervised by Dr. Amanda Parker. My research journey has been enriched by valuable experiences, including serving as a Research assistant at MIT Computational Connectomics Group led by Prof. Nir Shavit, where I had the privilege of collaborating with Prof. Hao Wang and Prof. Lu Mi. Additionally, I worked as a Research assistant at the National University of Singapore, supervised by Dr. Gang Guo.

Email  /  Google Scholar  /  Github

My research interest is to apply Interpretable Machine Learning to interdisciplinary fields.
In particluar, I am interested in:

• Interpretable ML (Concept Bottleneck Models, Rule-based Reasoning, Attribution-based Explanations, etc.)
• Large Language Models (Interpretability, Evaluation, Multimodal Reasoning, etc.)
• AI4Science (Healthcare, Environmental Science, Material Science, etc.)

"*" indicates equal contribution, "_" indicates equal advising.

We propose Concept-based Unsupervised Domain Adaptation, a framework that improves interpretability and robustness of Concept Bottleneck Models under domain shifts through adversarial training.

We introduce Energy-Based Concept Bottleneck Models (ECBMs) as a unified framework for concept-based prediction, concept correction, and fine-grained interpretations based on conditional probabilities.

We propose Energy-Based Conceptual Diffusion Models (ECDMs), a framework that unifies the concept-based generation, conditional interpretation, concept debugging, intervention, and imputation under the joint energy-based formulation.

Machine learning models were built to correlate the molecular composition and biological reactivity at the world's largest reservoir. Shorter incubations result in a broader range of molecules disappearing, with a greater contribution of stochasticity. Tuning the machine learning model contributes to yield additional interpretability beyond its well-recognized predictive power.

Medical image data are often limited due to the expensive acquisition and annotation process. Hence, training a deep-learning model with only raw data can easily lead to overfitting. To this end, we present a novel method, called Dynamic Data Augmentation (DDAug), which is efficient and has negligible computation cost.

Photochemical reactions are essential components altering dissolved organic matter (DOM) chemistry. We first used machine learning approaches to compatibly integrate existing irradiation experiments and provide novel insights into the estuarine DOM transformation processes.

Small subgraphs (graphlets) are important features to describe fundamental units of a large network. Unfortunately due to the inherent complexity of this task, most of the existing methods are computationally intensive and inefficient. In this work, we propose GNNS, a novel representational learning framework that utilizes graph neural networks to sample subgraphs efficiently for estimating their frequency distribution.

This thesis develops an enhanced version based on the original Iterative Label Spreading Clustering which is specially designed for materials science, and aim to become faster and have fewer hyperparameters.

In this paper, we present our solution for the ICASSP 2021 Network Anomaly Detection Challenge (NAD) challenge. Our approach ranked as 2nd place in the final leaderboard.

We use a variety of classification models to recognize the positive cases of SARS. We conduct evaluation with two types of SARS datasets, numerical and categorical types. For the sake of more clear interpretability, we also generate explanatory rules for the models.

• Hong Kong PhD Fellowship, 2022 - 2026
• Conference Travel Allowance of HKPFS, 2023 - 2025
• HKUST RedBird PhD Award, 2022 - 2023
• HKUST Academic Excellence Awards (Continuing Students), 2023 - 2025

• IEEE Graduate Student Member, since 2021
• Australian Computer Society Member, since 2021
• Asia Pacific Neural Network Society Member, since 2021
• ICONIP Program Committee Member, 2023 - 2025
• ICLR Reviewer, 2025 - 2026
• T-PAMI Reviewer
• Teaching Assistant of Statistics of Stochastic Process & Algorithms and Analysis, 2021 (SJTU Summer School)
• Teaching Assistant of ELEC 1200 A System View of Communications: from Signals to Packets, Spring & Fall 2023 (HKUST)

The source code is from Jon Barron's public academic website.