Kaleidoscope: Learnable Masks for Heterogeneous Multi-agent Reinforcement Learning

Authors: Xinran Li, Ling Pan, Jun Zhang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical evaluations across extensive environments, including multi-agent particle environment, multi-agent Mu Jo Co and Star Craft multi-agent challenge v2, demonstrate the superior performance of Kaleidoscope compared with existing parameter sharing approaches, showcasing its potential for performance enhancement in MARL.
Researcher Affiliation Academia Xinran Li Ling Pan Jun Zhang Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology xinran.li@connect.ust.hk, lingpan@ust.hk, eejzhang@ust.hk
Pseudocode No The paper does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block. The technical details are described through prose and mathematical formulas.
Open Source Code Yes The code is publicly available at https://github.com/LXXXXR/Kaleidoscope.
Open Datasets Yes We test our proposed Kaleidoscope on three benchmark tasks: MPE (Lowe et al., 2017), Ma Mu Jo Co (Peng et al., 2021) and SMACv2 (Ellis et al., 2024).
Dataset Splits No The paper mentions using multiple random seeds for statistical robustness and reporting 95% confidence intervals, but it does not specify explicit training/validation/test dataset splits with percentages or counts for a single dataset.
Hardware Specification Yes The experiments on the SMACv2 benchmark were conducted using NVIDIA Ge Force RTX 3090 GPUs, while the experiments on other benchmarks were performed using NVIDIA Ge Force RTX 3080 GPUs.
Software Dependencies No The paper mentions the use of existing codebases (HARL, EPy MARL, Py MARL2) but does not provide specific version numbers for software dependencies or libraries.
Experiment Setup Yes Hyperparameters To ensure a fair comparison, we implement our method and all the baselines using the same codebase with the same set of hyperparameters, with the exception of method-specific ones. The common hyperparameters are listed in Tables 3 to 5. The Kaleidoscope-specific hyperparameters are provided in Table 6.