icdm2025-conference-notes

🌟 Day 4 – Takeaways from ICDM 2025, Washington, DC 🌟

Today was my last day at ICDM 2025, and it was a great way to wrap up an intense and inspiring week.

Keynote III – Dr. Jure Leskovec (Stanford University)

Title: “Relational Foundation Models: A New Frontier for Predictive AI in Structured Data”

Dr. Leskovec emphasized that relational data is the most important and fascinating type of data.

Key points:

His ideas on relational foundation models also sparked several new thoughts for me on domain adaptation, which I plan to explore further in my own research.

10:00–12:00 – Poster Session

I spent the late morning at the Poster Display, where I presented my poster and had some thoughtful conversations about my work.

Afternoon – Session Chair, Undergraduate & High School Symposium

In the afternoon, I served as Session Chair for the Undergraduate and High School Symposium:

The research quality, professionalism, and depth of the Q&A discussions were all very impressive.

I genuinely felt I was seeing the future of science in these students. 🚀

16:00–17:30 – Foundations Session (Paper Presentation)

I also presented my paper in the Foundations session (only three papers in the slot).

A striking moment: when I mentioned the EM algorithm and asked for a show of hands, less than 30% of the audience was familiar with it.

This really reinforced a concern I’ve had for a while:

We must keep teaching our students fundamental algorithms and principles, not just how to tweak LLMs.
Without strong foundations, “foundation models” are sitting on shaky ground.

Closing Ceremony (17:40–18:30)

The day ended with the closing ceremony. It was announced that ICDM 2026 will be held in Shenyang, China.

Final Reflection

ICDM 2025 has been both intellectually rich and personally motivating.

I’ll share a few photos from my talk, poster session, and today’s events alongside this blog.

Thanks to all the organizers, speakers, and participants for a memorable conference. 🙏

Audience listening to a keynote on foundation models for structured data.
Keynote by Dr. Jure Leskovec on foundation models for structured data.
Slide showing examples of structured data models in commerce, finance, and social media.
Illustration of how predictions on structured data power everyday applications.
Slide explaining how predictions drive revenue across several application domains.
Emphasis on the economic impact of accurate predictions from structured data models.
Slide titled 'We are stuck with Machine Learning!' discussing current challenges.
Discussion of talent scarcity, high costs, and maintenance challenges in traditional ML pipelines.
Slide introducing relational foundation models via in-context learning.
Introduction of relational foundation models as a new paradigm for predictive AI on structured data.
Presenter standing beside a poster on EM-based transfer learning for Gaussian causal models.
Presenting my poster on EM-based transfer learning for Gaussian causal models under covariate and target shift.
Presenter and a colleague standing next to the EM-based transfer learning poster.
Discussing the poster and research contributions with conference attendees.
Presenter and another attendee in front of the EM-based transfer learning poster.
Further conversations at the poster session about domain adaptation and causal models.
Group photo of session chair with undergraduate and high school presenters.
Group photo with the undergraduate and high school symposium presenters after the session.
Large group photo of symposium participants and organizers.
Closing group picture with student researchers and their parents.