Xiangliang Zhang

This article examines security risks of backdoor attacks on machine learning in graph systems.

This study measures LLMs’ performance in complex math dialogues.

Introducing MoeSumm: a flexible and efficient summarization model.

MolX improves Large Language Models' grasp of molecular data and chemistry tasks.

Exploring vulnerabilities in Personalized Federated Learning and emerging backdoor attack methods.

A new method enhances language models by integrating knowledge across languages.

New methods improve accuracy in identifying animal sounds for wildlife monitoring.

Introducing ScholarChemQA, a dataset for chemical question answering to support researchers.

SaSR-Net connects sounds and visuals to accurately answer questions about videos.

Researchers find effective ways to remove unwanted knowledge from language models.

Researchers propose new methods to keep LLMs safe from harmful content generation.