About me

I am a PhD student in Data Science and Artificial Intelligence at Monash University, supervised by A/Prof. Zongyuan Ge and A/Prof. Dominic Dwyer. I develop learning and evaluation methods for AI systems that reason across modalities, operate with limited or imperfect evidence, and need to remain dependable in scientific and high-stakes clinical settings.

My current research connects multimodal language models for mental health, medical image analysis, and translational AI for clinical practice. Across these directions, I am particularly interested in process-level evaluation, data efficiency, safety, and clinical utility.

I received my M.Eng. in Electronic and Information Engineering from Shanghai Jiao Tong University in 2024, supervised by A/Prof. Lichi Zhang and Prof. Qian Wang of ShanghaiTech University, and was named a Shanghai Outstanding Graduate. I received my B.Eng. in Biomedical Engineering from Beihang University in 2021, graduating first in my cohort (1/64) and receiving the Beijing Outstanding Graduate honour.

Before beginning my PhD, under the supervision of Dr. Xi Ouyang and Prof. Dinggang Shen, I contributed at United Imaging Intelligence to developing a segmentation foundation model for whole-body CT imaging, designed to support broad anatomical coverage within a unified model. I later worked on multimodal AI for mental health at Monash Suzhou and then joined Xiaohongshu (RED) hi Lab, where I contributed to post-training the dots2 model for audio understanding, particularly for paralinguistic cues and emotionally engaging dialogue.

Outside research, I am an amateur photographer with ten years of experience behind the camera, documenting everyday life, travel, and people. I share selected photographs on Instagram.

Research

  • Multimodal LLMs for Mental Health

    Audio, vision, and language models for depression detection, safety and ethics evaluation, process analysis, and clinician-oriented explanation.

  • Medical Image Analysis

    Robust segmentation from scarce, partially labelled, synthetic, or noisy data, spanning task-specific networks and medical foundation models.

  • Translational AI for Clinical Practice

    Clinically grounded AI systems designed around unmet needs in diagnosis, decision support, and care pathways, with an emphasis on clinical utility, robust evaluation, and real-world translation.

News

  • AudioProcessBench was released.
  • PsychEthicsBench was accepted to Findings of ACL 2026.
  • CHiRPE was accepted to EACL 2026 Short Papers.
  • It Hears, It Sees Too was released.
  • Exploiting Latent Classes was accepted to MICCAI 2024.

Selected publications

Google Scholar

First-author publications only.

  1. AudioProcessBench: Benchmarking Process Errors in Audio-Grounded Reasoning

    Xiangyu Zhao et al. arXiv, 2026.

  2. AdLER: Adversarial Training with Label-Error Rectification for One-Shot Medical Image Segmentation

    Xiangyu Zhao et al. Expert Systems with Applications, 2026.

  3. It Hears, It Sees Too: Visual Understanding in Audio Language Models for Depression Detection

    Xiangyu Zhao et al. arXiv, 2025.

  4. Exploiting Latent Classes for Medical Image Segmentation from Partially Labeled Datasets

    Xiangyu Zhao et al. MICCAI, 2024.

  5. sTBI-GAN: An Adversarial Learning Approach for Data Synthesis on Traumatic Brain Segmentation

    Xiangyu Zhao et al. Computerized Medical Imaging and Graphics, 2024.

  6. RCPS: Rectified Contrastive Pseudo Supervision for Semi-Supervised Medical Image Segmentation

    Xiangyu Zhao et al. IEEE Journal of Biomedical and Health Informatics, 2023.

  7. One-Shot Traumatic Brain Segmentation with Adversarial Training and Uncertainty Rectification

    Xiangyu Zhao et al. MICCAI, 2023.

  8. Prior Attention Network for Multi-Lesion Segmentation in Medical Images

    Xiangyu Zhao et al. IEEE Transactions on Medical Imaging, 2022.

  9. D2A U-Net: Automatic Segmentation of COVID-19 CT Slices Based on Dual Attention and Hybrid Dilated Convolution

    Xiangyu Zhao et al. Computers in Biology and Medicine, 2021.

Experience

Xiaohongshu (RED)

Research Intern · Audio-understanding post-training for dots2, focusing on paralinguistic cues and emotionally engaging dialogue

Monash Suzhou

Research Assistant · Multimodal AI for mental health

Education

Beihang University

B.Eng. in Biomedical Engineering · Rank 1/64 · Beijing Outstanding Graduate

Contact

I’m always interested in thoughtful collaborations across multimodal AI, scientific discovery, and clinical research.

Google ScholarGitHubLinkedIn