Learning to Ground Multi-Agent Communication with Autoencoders
Authors: Toru Lin, Jacob Huh, Christopher Stauffer, Ser Nam Lim, Phillip Isola
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We experimentally validate that this is an effective approach for learning decentralized communication in MARL settings: a communication model trained with a simple autoencoder can consistently outperform baselines across various MARL environments. In this section, we demonstrate that autoencoding is a simple and effective representation learning algorithm to ground communication in MARL. We evaluate our method on various multi-agent environments and qualitatively show that our method outperforms baseline methods. We then provide further ablations and analyses on the learned communication. |
| Researcher Affiliation | Collaboration | Toru Lin MIT CSAIL torulk@mit.edu Minyoung Huh MIT CSAIL minhuh@mit.edu Chris Stauffer Facebook AI cstauffer@fb.com Ser-Nam Lim Facebook AI sernamlim@fb.com Phillip Isola MIT CSAIL phillipi@mit.edu |
| Pseudocode | No | The paper includes a schematic diagram (Figure 1) to illustrate the system's overview and describes the architecture verbally, but it does not provide any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | Project page, code, and videos can be found at https://toruowo.github.io/marl-ae-comm/. |
| Open Datasets | Yes | We design CIFAR Game following the setup of Multi-Step MNIST Game in [17], but with CIFAR-10 dataset [23] instead. |
| Dataset Splits | No | The paper reports evaluation metrics like '100 episodes per seed over 10 random seeds' and discusses training details in the Appendix, but it does not explicitly state dataset splits (e.g., 80/10/10) for training, validation, and testing. |
| Hardware Specification | No | The paper mentions 'The exact details of the network architecture and the corresponding training details are in the Appendix.' and the authors' checklist states they included compute resources. However, specific hardware details such as GPU models or CPU types are not present in the main body of the provided paper text. |
| Software Dependencies | No | The paper describes the use of deep neural networks, A3C, GRU, and MLPs, but it does not specify any software dependencies with version numbers (e.g., Python version, specific deep learning frameworks like PyTorch or TensorFlow, or library versions). |
| Experiment Setup | Yes | The exact details of the network architecture and the corresponding training details are in the Appendix. |