PHYRE

PHYRE是一个专为测试和提升机器在2D物理环境中的推理能力而设计的基准数据集。该数据集包含一系列简单的经典力学谜题，旨在鼓励开发样本效率高且能跨谜题泛化的学习算法。PHYRE通过其独特的任务设计，强调物理推理、泛化能力和样本效率，为评估和提升人工智能在物理理解方面的能力提供了平台。数据集中的每个谜题都有一个目标状态和一个初始状态，通过在环境中放置一个或多个新物体来解决问题。PHYRE不仅适用于机器学习研究，也适用于教育领域，帮助学生和研究人员理解物理现象和机器学习算法的交互。

PHYRE is a benchmark dataset specifically designed for testing and enhancing machines' reasoning capabilities in 2D physical environments. This dataset contains a series of simple classical mechanics puzzles, aiming to spur the development of learning algorithms with high sample efficiency and strong cross-puzzle generalization ability. Through its unique task design, PHYRE emphasizes physical reasoning, generalization performance and sample efficiency, providing a platform for evaluating and improving artificial intelligence's capacity for physical understanding. Each puzzle in the dataset has a target state and an initial state, and the problem can be solved by placing one or more new objects into the environment. PHYRE is applicable not only to machine learning research, but also to the field of education, helping students and researchers understand physical phenomena and the interaction between machine learning algorithms.