Compositional Reasoning Under Uncertainty (CURI)

CURI数据集是由Facebook AI Research创建，用于评估机器学习系统在不确定条件下的组合推理能力。该数据集包含14,929个概念，涵盖了抽象理解、生产性泛化、布尔运算学习、变量绑定等多个方面，旨在模拟人类在无限概念空间中的学习和推理能力。CURI通过一系列元学习任务，评估模型在处理复杂概念和不确定性时的表现，推动了机器学习在组合性推理方面的研究进展。

The CURI dataset was created by Facebook AI Research to evaluate the compositional reasoning capabilities of machine learning systems under uncertain conditions. It contains 14,929 concepts covering multiple aspects such as abstract understanding, productive generalization, Boolean operation learning, variable binding and more, aiming to simulate human learning and reasoning abilities within an infinite conceptual space. Through a series of meta-learning tasks, CURI assesses model performance when handling complex concepts and uncertainties, advancing research progress in compositional reasoning for machine learning.