Taewon Kim

PhD @ KAIST AI, advised by Prof. Sungsoo Ahn. Building foundation models for biology — proteins, genomes, molecules, and scientific discovery.

[Minkyu Kim @minkyu1022]

🎉 "CatFlow: Co-generation of Slab-Adsorbate Systems via Flow Matching" has been accepted at #ICML2026!

We develop flow matching for catalyst design at the all-atom level with a factorized representation.

Thanks to @__na__young__ , @honghui_kim , and @sungsoo_ahn_

[Taewon Kim @taewonKKK]

Huge thanks to Hyosoon Jang (co-first) and @sungsoo_ahn_ for their support and guidance.

This is part of the bigger “K-Fold” project, supported by the Korean Ministry of Science and ICT.

📄 Paper: https://t.co/ANrlAWZ1lK
💻 Code coming soon. Stay tuned! (9/9)

[Taewon Kim @taewonKKK]

What should a structure-aware protein representation be good at?

It should provide geometric signals that structure models can use.

We introduce:
1️⃣ TriProRep: atom-level protein representations
2️⃣ REPSP: a structure-predictive benchmark

https://t.co/pFv2AdLkuM 👇 (1/9)

[Taewon Kim @taewonKKK]

Finally, REPSP tests whether representations can supervise folding models directly, as in REPA.

TriProRep works not only as input features for co-folding, but also as a representation-alignment target for monomer structure prediction. (8/9)

[Taesik Gong @Taesik_MobileAI]

I am thrilled to share that our work about test-time adaptation has been accepted to #NeurIPS2022!
Many thanks to my collaborators: @jongheon_j, Taewon Kim, @hai_yewon, @jinwoos0417, @wewantsj
Code and camera-ready version will be available soon.
Stay tuned! @NeurIPSConf