Yichao Cai
Ph.D. Candidate
AIML, Adelaide University
About Me

I am a third-year Ph.D. student in Computer Science at the Australian Institute for Machine Learning (AIML), Adelaide University (formerly The University of Adelaide), advised by Prof. Javen Qinfeng Shi. I study representation learning, asking how natural language supervision determines which semantic structure is captured, preserved, or lost in vision-language models

My work combines theoretical analysis with controlled empirical study to understand contrastive learning, cross-modal alignment, and identifiability in representation learning, with the long-term goal of building more interpretable and reliable multimodal AI systems.


News
2026
We had three papers on representation learning accepted to ICML 2026: one first-author paper on contrastive learning geometry and two co-authored papers on AI4Science and graphical modeling, respectively.
May 01
Jan 28
I attended MLSS Melbourne 2026 and enjoyed learning from world-class speakers and connecting with the community.
Jan 26
2025
I served as a guest lecturer in Statistical Machine Learning and presented recent advances in vision-language modeling. Slides.
Oct 15
Education
  • Sep 2023 - Present
    Ph.D. in Computer Science, Adelaide University
    Adelaide University
    Advisor: Prof. Javen Qinfeng Shi
  • Sep 2016 - Jun 2019
    M.Sc. in Instrument Science and Technology, Wuhan University of Technology
    Wuhan University of Technology
    Advisor: Prof. Xiao Zhou
  • Sep 2012 - Jun 2016
    B.Eng. in Measurement & Control Technology and Instrument, Wuhan University of Technology
    Wuhan University of Technology
Experience
  • Sep 2023 - Present
    PhD Student Researcher, AIML, Adelaide University
  • Jun 2020 - Aug 2022
    AI Engineer, Tellhow Software
  • Jul 2019 - Apr 2020
    Software Engineer, Huawei Technologies
  • May 2018 - Oct 2018
    Visiting Student Researcher, California PATH, UC Berkeley
Teaching
  • Teaching Assistant - Neural Networks and Deep Learning (ARTI X300), Adelaide University
    Semester 1 2026
  • Guest Lecturer and Head Tutor - Statistical Machine Learning (COMP SCI 3314), Adelaide University
    Semester 2 2025
  • Teaching Assistant - Using Machine Learning Tools (COMP SCI 7317), Adelaide University
    Trimester 2 2025
  • Teaching Assistant - Concepts in AI and ML (COMP SCI 7327), Adelaide University
    Semester 1 2025
Honors & Awards
  • NeurIPS Scholar Award
    2025
  • Adelaide University Research Scholarships
    2023
  • Award for Outstanding Graduates, Wuhan University of Technology
    2019
Academic Service
  • Reviewer for TMLR, ICLR 2026, ICML 2026, and NeurIPS 2026
Selected Publications (view all )
The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-Modal Divergence
ICML
The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-Modal Divergence

Yichao Cai; Zhen Zhang; Yuhang Liu; Javen Q. Shi.

International Conference on Machine Learning (ICML) 2026

A theoretical study of representation geometry in unimodal and multimodal contrastive learning, revealing a geometric bifurcation and identifying object-level causes of the modality gap.

[Paper] [Project] [Code]

The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-Modal Divergence

Yichao Cai; Zhen Zhang; Yuhang Liu; Javen Q. Shi.

A theoretical study of representation geometry in unimodal and multimodal contrastive learning, revealing a geometric bifurcation and identifying object-level causes of the modality gap.

[Paper] [Project] [Code]

ICML
I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?
ICLR
I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?

Yuhang Liu; Dong Gong; Yichao Cai; Erdun Gao; Zhen Zhang; Biwei Huang; Mingming Gong; Anton van den Hengel; Javen Q. Shi.

International Conference on Learning Representations (ICLR) 2026

An investigation of whether next-token prediction alone is sufficient for learning human-interpretable concepts from data.

[Paper] [Project]

I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?

Yuhang Liu; Dong Gong; Yichao Cai; Erdun Gao; Zhen Zhang; Biwei Huang; Mingming Gong; Anton van den Hengel; Javen Q. Shi.

An investigation of whether next-token prediction alone is sufficient for learning human-interpretable concepts from data.

[Paper] [Project]

ICLR
On the Value of Cross-Modal Misalignment in Multimodal Representation Learning
NeurIPS
On the Value of Cross-Modal Misalignment in Multimodal Representation Learning

Yichao Cai*; Yuhang Liu*; Erdun Gao; Tianjiao Jiang; Zhen Zhang; Anton van den Hengel; Javen Q. Shi. (* equal contribution)

Advances in Neural Information Processing Systems (NeurIPS) 2025 Spotlight

Studies when controlled cross-modal misalignment can improve multimodal representation learning instead of only harming it.

[Paper] [Project] [Code]

On the Value of Cross-Modal Misalignment in Multimodal Representation Learning

Yichao Cai*; Yuhang Liu*; Erdun Gao; Tianjiao Jiang; Zhen Zhang; Anton van den Hengel; Javen Q. Shi. (* equal contribution)

Spotlight

Studies when controlled cross-modal misalignment can improve multimodal representation learning instead of only harming it.

[Paper] [Project] [Code]

NeurIPS
CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts
ECCV
CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts

Yichao Cai; Yuhang Liu; Zhen Zhang; Javen Q. Shi.

European Conference on Computer Vision (ECCV) 2024

Explores language-guided disentanglement of style and content through contrastive learning with augmented prompts.

[Paper] [Project] [Code]

CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts

Yichao Cai; Yuhang Liu; Zhen Zhang; Javen Q. Shi.

Explores language-guided disentanglement of style and content through contrastive learning with augmented prompts.

[Paper] [Project] [Code]

ECCV