← Search

Haoyun Xu

2 accepted papers

2026

The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMs

ICLR 2026poster

Diffusion-based large language models (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs, offering faster inference and greater interactivity via parallel decoding and bidirectional modeling. However, despite strong performance in code generation and text infilling, we i…

Cited by 0SourcecodeScholar
2025

Let’s Focus on Neuron: Neuron-Level Supervised Fine-tuning for Large Language Model

COLING 2025main

Large Language Models (LLMs) are composed of neurons that exhibit various behaviors and roles, which become increasingly diversified as models scale. Recent studies have revealed that not all neurons are active across different datasets, and this sparsity correlates positively with the task-specific…