InDL

InDL数据集是由爱丁堡大学信息学院创建，专注于通过视觉错觉评估深度学习模型在图表逻辑解释中的能力。该数据集包含10,000个样本，利用六种经典几何视觉错觉设计，旨在严格测试和基准化这些模型。数据集的创建过程涉及心理学原理的应用，通过比较人类和机器的视觉感知来量化模型的逻辑理解能力。InDL数据集的应用领域主要集中在改进深度学习模型在逻辑解释任务中的性能，解决模型在复杂逻辑推理中的‘黑箱’问题。The InDL dataset was developed by the School of Informatics at the University of Edinburgh, with the primary objective of evaluating deep learning models’ capabilities in logical diagram interpretation using visual illusions. Comprising 10,000 samples, this dataset is constructed based on six classic geometric visual illusions, aiming to rigorously test and benchmark these models. The construction process of the dataset applies psychological principles, quantifying the models’ logical comprehension capacities by comparing visual perceptions between human beings and machines. The main application scenarios of the InDL dataset focus on enhancing the performance of deep learning models in logical interpretation tasks, and addressing the "black box" problem of models in complex logical reasoning.

搜集汇总

背景与挑战

背景概述

InDL是一个基于视觉错觉的“图中逻辑”解释评估数据集，旨在研究深度学习模型处理感知与逻辑复杂交互的能力，以揭示模型的“黑箱”特性。该数据集包含数据生成脚本、多个基准测试子集（如dataset01至dataset05）以及用于模型训练和结果分析的Jupyter笔记本，支持相关论文的复现和研究。