A research paper authored by students Fang Zitao, Yu Shuyang, Guo Yifei, Zhang Yiwei, Cao Yiyao, and Dr. Sim Kuan Goh from the School of Artificial Intelligence and Robotics at Xiamen University Malaysia (XMUM) has been accepted for presentation at the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025).
Overview of EMNLP
EMNLP is acknowledged as a leading international conference in the field of Natural Language Processing (NLP) and holds an A* ranking according to the Computing Research and Education (CORE) Conference Ranking system.
Research Focus
The paper, titled “To See a World in a Spark of Neuron: Disentangling Multi-task Interference for Training-free Model Merging,” investigates the functions of neurons with respect to fine-tuning applications across domains such as vision, language, and large language models (LLMs). This study presents a novel approach named NeuroMerging, which seeks to integrate multiple AI models with varying strengths into a unified multi-task model. The method is particularly significant due to the growing accessibility of publicly available AI models, allowing for the leveraging of these models’ capabilities without incurring additional training or inference costs.
Contributors
Fang Zitao, a final-year undergraduate student in the Artificial Intelligence Programme, leads the research initiative, with Dr. Sim Kuan Goh acting as the supervising faculty member. Yu Shuyang and Guo Yifei, who are also part of the Class of 2025, are pursuing their Master’s degrees at Columbia University and Duke University, respectively. The research benefitted from contributions by Guodong Du, Jing Li, and Ho-Kin Tang from the Hong Kong Polytechnic University and the Harbin Institute of Technology, Shenzhen.
Funding Support
Funding for this project came from several organizations, including the Ministry of Higher Education Malaysia through the Fundamental Research Grant Scheme, the National Science Foundation of China, and the Shenzhen Science and Technology Program, among others.
Further Information
For further details, a preprint of the article is available on ArXiv. Additional information regarding the project can be found on the project’s webpage.
(Source: Xiamen University Malaysia)
