Seok-Ju Hahn

(Suck-Zoo)

Hi, I’m Seok-Ju! I am a postdoctoral appointee in the Mathematics and Computer Science Division at Argonne National Laboratory.

I build adaptive decision layers that help learning systems determine how and where to spend computation and when to trust a combination of knowledge, with guarantees. Especially, this includes federated and decentralized settings where data distributions, computational resources, communication patterns, and participant objectives are heterogeneous.

In my free time, I sing 🎤 R&B and do kettlebell 🐮🔔 swings. I am open to new collaborations and opportunities. Let’s connect!

Research Highlights

Personalized federated learning. Developed SuPerFed, which connects global and local models through low-loss subspaces for personalization under statistical heterogeneity (KDD 2022).
Adaptive decision-making for collaboration. Developed AAggFF, which unified and reframed long-term performance fairness in federated learning as online optimization over aggregation weights (ICML 2024).
Privacy-preserving data collaboration. Developed Diffusion Federated Datasets, a cooperative sampling framework that composes pretrained local diffusion models without exchanging model parameters and supports optional differential privacy (NeurIPS 2025).

sjhahn11512 [at] gmail [dot] com
hahns [at] anl [dot] gov

I am currently open to roles as a Postdoctoral Researcher, (Research) Assistant Professor, and ML Research Scientist.