The 15th ChinaGraph conference took place from October 10 to 13, 2024. Ollie Woodman, a student at the State Key Lab of CAD&CG, College of Computer Science and Technology, Zhejiang University, along with Professor Chen Wei, authored a paper that stood out among 258 submissions. Their work, titled "Exploring the neural landscape: Visual analytics of neuron activation in large language models with NeuronautLLM," not only secured acceptance but also claimed the conference's Best Paper Award in the English category.
Academician Hu Shimin presented the prestigious award to Ollie Woodman
About the Conference
ChinaGraph, established in 1996, is jointly sponsored by the China Computer Federation, Chinese Association of Automation, China Graphics Society, China Society of Image and Graphics, China Simulation Federation, and Hong Kong Society for Multimedia and Image Computing. It stands as the premier academic conference for computer graphics in China, fostering crucial exchanges among Chinese scholars in the field. Held biennially, the conference has successfully completed 14 sessions to date. The University of Science and Technology of China and Huangshan University co-hosted the 15th ChinaGraph 2024.
Paper Overview
Authors: Ollie Woodman, Chen Wei, et al.
Large language models (LLMs) such as OpenAI's ChatGPT and Google's Gemini have been pivotal in recent artificial intelligence and machine learning advancements. However, the immense size and dimensionality of these models pose significant challenges for interpretation and visualization. NeuronautLLM introduces a visual analysis system that identifies and visualizes neurons affecting the Transformer language model under user-defined prompts.
The paper's methodology combines straightforward yet informative visualizations with neuronal interpretation and classification data, opening up rich avenues for future research. Two experts reviewed NeuronautLLM, confirming its efficacy as a practical model interpretation tool. Interviews with five LLM specialists and usability tests demonstrate NeuronautLLM's exceptional user-friendliness and readiness for real-world applications. Furthermore, two in-depth case studies showcase NeuronautLLM's versatility in analyzing a wide range of LLM research questions.
