Kyomin Jung

Erase or Hide? Suppressing Spurious Unlearning Neurons for Robust Unlearning

PEARL: Differentially Private and Entropy-Aware Regulated Language Generation

Online Budget-Aware Guidance for Blockwise Discrete Text Diffusion

Training Dynamics of Parametric and In-Context Knowledge Utilization in Language Models

Program Synthesis via Test-Time Transduction

Guaranteed Generation from Large Language Models

Knowledge-localized Unlearning for Faithful Forgetting in Language Models