VNA Benchmark
收藏资源简介:
VNA Benchmark是由卡内基梅隆大学创建的数据集,专注于评估大型视觉语言模型在基础视觉网络分析任务上的零样本能力。该数据集包含100个图表,涵盖了识别图表中的最大度节点、判断三元组的结构平衡性以及计数图表中的组件等任务。数据集的创建旨在通过这些任务测试模型对图论概念的理解和应用能力,特别是在无需预先训练的情况下。此外,数据集的应用领域包括但不限于网络科学、机器学习和人工智能,旨在解决模型在处理复杂图形数据时的准确性和效率问题。
VNA Benchmark is a dataset developed by Carnegie Mellon University, focusing on evaluating the zero-shot capabilities of large vision-language models on fundamental visual network analysis tasks. This dataset comprises 100 graphs, covering tasks including identifying nodes with the maximum degree in a graph, assessing the structural balance of graph triples, and counting connected components within graphs. The dataset is designed to test models' understanding and application of graph-theoretic concepts, particularly in zero-shot scenarios without prior fine-tuning. Furthermore, the dataset's applicable domains include but are not limited to network science, machine learning, and artificial intelligence, aiming to address the accuracy and efficiency challenges faced by models when processing complex graph data.




