Xiangyuan Xue
BSc (Hons) in Computer Science at The University of Auckland (Expected 2028).
Research interests: on-device small language models, wearable sensing foundation models,
and multimodal/embodied reasoning.
About
I am an undergraduate Computer Science (Honours) student at the University of Auckland. I work on efficient and practical machine learning systems, with a focus on building and evaluating small language models for mental health applications and foundation models for wearable sensor data. I am also interested in multimodal reasoning and embodied agents that plan and act in long-horizon tasks.
Publications
-
Menta: A Small Language Model for On-Device Mental Health Prediction
Research Interests
- SLM Foundation Models for Wearable Sensors: building SLM-style foundation models for wearable sensor data and studying generalizable representations from large-scale physiological and activity datasets.
- Multimodal Reasoning Large Language Models (MLLMs): investigating planning–execute–reflect style loops for long-horizon tasks, with a focus on adaptive reasoning horizons, feedback-based rewards, and reflection-driven re-planning.
- On-Device Mental Health Prediction: fine-tuning and evaluation of a small language model for mental health-related prediction, including zero-shot, few-shot, and fine-tuning experiments, plus deployment experiments on mobile devices.
Education
Contact
Email: xxue752@aucklanduni.ac.nz
GitHub: https://github.com/xxue752-nz