Reasoning Riddles
收藏资源简介:
该数据集名为“推理谜题”,由马萨诸塞大学阿默斯特分校的研究团队创建,包含221个跨越六个认知类别的重新谜题。这些谜题旨在探究视觉语言模型在复杂推理场景中的解释能力,数据集包含丰富的视觉元素和文本线索,需要模型将多个认知技能整合以生成抽象的解决方案。该数据集是专门为解释性研究设计的,谜题按认知维度进行了分类,可用于评估和改进视觉语言模型的推理能力和透明度。
This dataset, named "Reasoning Puzzles", was created by a research team from the University of Massachusetts Amherst. It comprises 221 reframed puzzles spanning six cognitive categories. These puzzles are designed to investigate the explanatory capabilities of vision-language models (VLMs) in complex reasoning scenarios. The dataset features rich visual elements and textual cues, requiring models to integrate multiple cognitive skills to generate abstract solutions. Specifically tailored for explanatory research, the puzzles are categorized along cognitive dimensions, which can be used to evaluate and enhance the reasoning abilities and transparency of vision-language models.




