MABIM (Multi-Agent Benchmark for Inventory Management)
收藏资源简介:
MABIM是由微软亚洲研究院开发的多代理强化学习基准,专注于库存管理问题。该数据集模拟了一个多层级、多商品的库存管理环境,能够生成具有不同挑战性质的多样化任务。MABIM基于OpenAI Gym框架构建,旨在促进库存管理领域的研究进展,并提供一个平台来评估MARL算法在各种任务中的性能。数据集包含超过2000个真实需求数据,支持多层级仓库管理,并处理大量商品。MABIM不仅为解决库存管理挑战提供了一个开放和有效的基准,还通过其灵活性模拟了一系列MARL挑战,如规模扩展、合作、竞争、泛化和鲁棒性,进一步增强了其在各种场景中的应用性。
MABIM is a multi-agent reinforcement learning benchmark developed by Microsoft Research Asia, focusing on inventory management issues. This dataset simulates a multi-echelon, multi-product inventory management environment, capable of generating diverse tasks with varying challenge characteristics. Built on the OpenAI Gym framework, MABIM aims to advance research in the inventory management domain and provide a platform for evaluating the performance of multi-agent reinforcement learning (MARL) algorithms across various tasks. The dataset contains over 2,000 real demand data points, supports multi-echelon warehouse management, and handles a large volume of products. Not only does MABIM provide an open and effective benchmark for resolving inventory management challenges, but it also simulates a range of MARL challenges via its flexibility, such as scalability, cooperation, competition, generalization, and robustness, further enhancing its applicability across diverse scenarios.

- 1A Versatile Multi-Agent Reinforcement Learning Benchmark for Inventory Management微软亚洲研究院 · 2023年



