Learning Efficient Multi-agent Communication: An Information Bottleneck Approach

Authors: Rundong Wang, Xu He, Runsheng Yu, Wei Qiu, Bo An, Zinovi Rabinovich

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To demonstrate the efficiency of our method, we conduct extensive experiments in various cooperative and competitive multi-agent tasks with different numbers of agents and different bandwidths.
Researcher Affiliation Academia 1School of Computer Science and Engineering, Nanyang Technological University, Singapore.
Pseudocode Yes Algorithm 1: Informative Multi-agent Communication
Open Source Code Yes 1Our code is provided in https://github.com/EC2EZ4RD/IMAC
Open Datasets Yes We evaluate IMAC on a variety of tasks and environments: cooperative navigation and predator prey in Multi Particle Environment (Lowe et al., 2017), as well as 3m and 8m in Star Craft II (Samvelyan et al., 2019).
Dataset Splits No The paper evaluates agents in simulated environments where data is generated during interaction, rather than using predefined dataset splits. No explicit train/validation/test splits are mentioned for a fixed dataset.
Hardware Specification No The paper does not provide specific details about the hardware used, such as GPU/CPU models or memory specifications.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes We first respectively train IMAC with different prior distributions z(Mi) of N(0, 1), N(0, 5), and N(0, 10), to satisfy different default limited bandwidth constraints. We train our agents by self-play for 100,000 episodes. Ablation studies also provide details on parameters like β.