Gyubin Lee

Hi! I'm a first-year MS student in the School of Computing at KAIST, under the supervision of Prof. Sungjin Ahn at the Machine Learning and Mind Lab (MLML). My research interests include Generative models, unsupervised Reinforcement Learning, and World Model.

Email  /  Google Scholar  /  Twitter  /  LinkedIn

Publications

We propose the Neural Theorizer (NEO), a learning paradigm where understanding emerges through building internal theories rather than merely predicting the future. NEO induces latent programs as a learned Language of Thought and executes them through a shared transition model, enabling explanation-driven generalization.

We introduce adaptive inference-time scaling-dynamically adjusting computational effort during inference-and propose Adaptive Bi-directional Cyclic Diffusion (ABCD), a flexible, search-based inference framework.

Projects

We tried to decompose the temporal latent variable into primitive components using compositional energy functions.

Education

KAIST (Sep 2025 - present)

MS student at the School of Computing (working with Sungjin Ahn)

Korea University (Mar 2019 - Aug 2025)

BS in Computer Science and Engineering (with Public Governance and Leadership Convergence Track)

Experience

Research Intern (Jan 2025 - Sep 2025) @ KAIST

on inference-time scaling methods for diffusion models

Research Intern (Apr 2024 - Sep 2024) @ Korea University

on Reinforcement Learning based post-training methods

Services

Reviews

Conferences: ICML (2025, 2026), ICLR (2026)

This page was built based on This Template