KANDY
收藏arXiv2024-02-27 更新2024-06-21 收录
下载链接：
https://github.com/continual-nesy/KANDYBenchmark
下载链接
链接失效反馈官方服务：
资源简介：
KANDY数据集是一个用于增量神经符号学习和推理的基准框架，专注于具有递增复杂性的二元分类任务。它包含两个任务课程，一个较简单，一个较难，旨在挑战神经和符号方法。数据集通过提供不同复杂度的任务，以及从完全监督到稀疏监督的监督策略，来评估持续学习和半监督学习方法。此外，KANDY还允许用户通过提供的工具生成自己的基准，从而扩展其应用范围。
The KANDY dataset is a benchmark framework for incremental neuro-symbolic learning and reasoning, focusing on binary classification tasks with increasing complexity. It consists of two task curricula: one simpler and the other more challenging, aimed at testing both neural and symbolic methods. The dataset evaluates continual learning and semi-supervised learning approaches by providing tasks of varying complexities, alongside supervision strategies ranging from fully supervised to sparse supervision. Additionally, KANDY allows users to generate custom benchmarks via the provided tools, thereby expanding its applicable scope.
提供机构：
University of Pisa, University of Modena and Reggio Emilia, University of Siena
创建时间：
2024-02-27
