SYGAR
收藏资源简介:
SYGAR数据集是由慕尼黑大学信息与语言处理中心创建的,旨在评估模型在抽象空间推理中进行系统性泛化的能力。该数据集包含了对二维对象进行的基本几何变换(如平移、旋转、反射)及其组合。通过在网格环境中对对象应用这些变换,数据集为研究提供了评估模型泛化能力的手段,特别是在处理之前未见过的变换组合时。SYGAR的创建是为了解决模型在抽象空间推理领域的系统性泛化问题,推动研究向更稳健、更通用的模型发展。
The SYGAR dataset was developed by the Center for Information and Language Processing at LMU Munich (Ludwig-Maximilians-Universität München) to evaluate models' systematic generalization capabilities in abstract spatial reasoning. This dataset covers basic geometric transformations (e.g., translation, rotation, reflection) applied to two-dimensional (2D) objects, as well as combinations of these transformations. By applying these transformations to objects within a grid-based environment, the dataset provides a research tool for assessing models' generalization performance, particularly when handling previously unseen transformation combinations. The SYGAR dataset was created to address the systematic generalization problem of models in the field of abstract spatial reasoning, and to advance research toward more robust and general-purpose models.




