Hokoff
收藏资源简介:
Hokoff数据集由清华大学、腾讯天美工作室和腾讯AI实验室联合创建,源自全球热门的多人在线战斗竞技场(MOBA)游戏《王者荣耀》。该数据集包含丰富的预训练模型生成的多样化数据,适用于离线强化学习和离线多智能体强化学习。数据集的创建过程利用了游戏中的多级别模型进行数据采样,确保了数据的真实性和复杂性。Hokoff数据集主要应用于强化学习算法的开发和评估,旨在解决现实世界中的复杂决策问题和多任务学习挑战。
The Hokoff dataset was jointly created by Tsinghua University, Tencent TiMi Studio Group, and Tencent AI Lab, originating from *Honor of Kings*, a globally popular multiplayer online battle arena (MOBA) game. It encompasses diverse data generated by a range of pre-trained models, and is tailored for offline reinforcement learning and offline multi-agent reinforcement learning. The dataset creation process utilized multi-level in-game models for data sampling, ensuring the authenticity and complexity of the collected data. Primarily applied to the development and evaluation of reinforcement learning algorithms, the Hokoff dataset aims to address complex real-world decision-making problems and multi-task learning challenges.




