2025 KIAS Winter School on Mathematics and AI

Temporary information page; official website coming soon.

TL;DR

Dates: December 2–5, 2025
Venue: Park Roche, Jeongseon — https://park-roche.com/
Apply Now: https://forms.gle/oaApy67SgbAsRrYMA (Google Form)

Home

Home

The 2025 KIAS Winter School on Mathematics and AI is a four‑day hackathon‑style program at the intersection of mathematics and AI. Teams work in parallel on hands‑on projects in math‑for‑AI, AI‑for‑math, and formalization. The program also includes invited talks on related topics.

Topics & Keywords (tentative)

Formalization (Lean), Autoformalization, Informalization
Machine Learning for Mathematics
Large Language Models
Reinforcement Learning

Organizers

Joonhyun La (KIAS)
Chul‑hee Lee (KIAS)
Kyu‑Hwan Lee (University of Connecticut / KIAS)

Team Leads (tentative)

Ilkyoo Choi (HUFS / DIMAG, IBS)
Byung-Hak Hwang (KIAS)
Jihoon Hyun (KAIST)
Chul‑hee Lee (KIAS)
Seewoo Lee (UC Berkeley)
Hyojae Lim (RICAM)

Speakers

Dohyun Kwon (University of Seoul)
Kyu‑Hwan Lee (University of Connecticut / KIAS)
Hongseok Yang (KAIST)

Contact

Email: chlee@kias.re.kr (Chul‑hee Lee)

Supported by

Korea Institute for Advanced Study (KIAS), HCMC

Program

Talks

Talk titles and abstracts TBA.

Dohyun Kwon (University of Seoul)

Title: TBA
Abstract: TBA

Kyu‑Hwan Lee (University of Connecticut / KIAS)

Title: TBA
Abstract: TBA

Hongseok Yang (KAIST)

Title: Tackling asymptotic extremal problems in graph theory using neural networks
Abstract: Assisting mathematical discovery with machine learning techniques has been an active research topic in the machine-learning community in recent years, yielding impressive results such as the discovery of key insights into or counterexamples for open conjectures in knot theory, representation theory, arithmetic geometry and combinatorics. In this talk, I will describe our ongoing efforts for extending this line of research to extremal combinatorics. We consider the asymptotic extremal problems on graphs, which are just a particular type of optimisation problems on an infinite limiting version of graphs, called graphons. We aim at helping prove or disprove open conjectures for such problems using tools from machine learning. Our idea is to represent these infinite limiting graphs using carefully-designed neural networks, and to solve these optimisation problems using gradient descent. I will describe what challenges we encountered, how we overcame those challenges by designing a particular architecture for neural networks inspired by the popular diffusion model, and what new insights into the well-known asymptotic extremal problems we gained by our approach. This is joint work with Taeyoung Kim from KAIST, Jineon Baek from KIAS, and Joonkyung Lee from Yonsei University.

Team Programs

Participants will be grouped into teams; the application page currently lists only the titles of team project topics, and more detailed information will be provided later: https://forms.gle/oaApy67SgbAsRrYMA.

Timetable

High‑level schedule; detailed times TBA.

Day 1 (Tue): Arrival & Kickoff

Arrival/registration/lunch; opening + team kickoff; evening hacking.

Day 2 (Wed):

Talk 1; Team hacking.

Day 3 (Thu):

Talk 2; Team hacking; Team presentations; banquet and networking.

Day 4 (Fri): Closing & Departure

Talk 3; lunch; departure.

Venue

Park Roche, Jeongseon — https://park-roche.com/

Transportation

Shuttle Bus (tentative)
- A shuttle bus will be arranged between KIAS and Park Roche for participants.
- KIAS → Park Roche: Dec 2 (Mon) 9:00 AM (approx. 3 h 30 min)
- Park Roche → KIAS: Dec 5 (Thu) 1:30 PM (approx. 3 h 30 min)
- Times are subject to change.

Registration

Apply Now: https://forms.gle/oaApy67SgbAsRrYMA (Google Form)