Xianghao Kong

Hello/你好👋

prof_pic.jpeg

I’m the last-year CS Ph.D. working with AMAZING Prof. ✨ Greg Ver Steeg ✨ at the University of California, Riverside.

My research focuses on Generative Models (Diffusion Models & Energy-Based Models), particularly their interpretability, alignment, and compositionality. Specifically, we interpret diffusion models from a novel information theory perspective, i.e., Information-Theoretic Diffusion (ITD). Our findings show that Pointwise Mutual Information (PMI) enhances compositional understanding and modality alignment (e.g. text and image). We’re continuing to expand the ITD universe 🌌 and are actively seeking collaborators! Feel free to reach out! 🚀

Before joining UCR, my focus was on EEG data analysis within Brain-Computer Interface (BCI) 🧠 technology, merging neuroscience with computer science. I’m a foodie living at SoCal and love sketching 🎨🖌 and visiting museums 🏛️.

news

Mar 01, 2025 I am honored to receive the Dissertation Completion Fellowship Award from UCR! 🎉
Feb 26, 2025 The paper Stretching Each Dollar: Diffusion Training from Scratch on a Micro-Budget has been accepted to CVPR 2025! Looking forward to sharing our work with the community! 🚀
Dec 05, 2024 Thrilled to share a new milestone in my academic journey: I successfully passed my dissertation proposal exam!
Oct 02, 2024 I have been invited to the Advancing AI 2024 event by AMD on October 10th ✈️. If you’ll be in San Francisco, I’d love to grab a coffee ☕️ or explore some great food together!
Sep 25, 2024 Time has flown by! This summer at Adobe, I had the pleasure of meeting so many wonderful people, and I will miss you all. However, my journey with Adobe isn’t over yet. I’m excited to continue working with Adobe Firefly to push the boundaries of diffusion models.

selected publications

  1. CVPR
    Stretching Each Dollar: Diffusion Training from Scratch on a Micro-Budget
    Vikash Sehwag , Xianghao Kong , Jingtao Li , and 2 more authors
    2025
  2. ICLR
    Interpretable Diffusion via Information Decomposition
    Xianghao Kong* , Ollie Liu* , Dani Yogatama , and 1 more author
    2024
  3. ICLR
    Information-Theoretic Diffusion
    Xianghao Kong , Rob Brekelmans , and Greg Ver Steeg
    2023
  4. ACL
    Asymmetric Bias in Text-to-Image Generation with Adversarial Attacks
    Haz Sameen Shahgir , Xianghao Kong , Greg Ver Steeg , and 1 more author
    2024