Emergent communication at scale

This repo allows exploring different dynamics and RL algorithms to train a population of agents that communicate to solve a referential game. It is based on the Jax/Haiku python api, and the Jaxline training framework, smoothly allowing for large-scale and multi-device experiments. The repo contains the code which we used for the experiments reported in our ICLR 2022 paper, titled "Emergent Communication At Scale".

Downloading dataset

Download the datasets CelebA logits and ImageNet logits in the emcom_datasets/ directory.

Installation

Install dependencies by running the following commands:

# Create a virtual env
python3 -m venv ~/.venv/emcom
# Switch to the virtual env
source ~/.venv/emcom/bin/activate
# Install other dependencies
pip install -r requirements.txt
# Manually install the latest Jaxline version, since required 0.0.6 is not yet on pypi.
pip install git+https://github.com/deepmind/jaxline

Lewis game

To run the Lewis game experiment The Lewis game experiment trains a population of agents and computes different metrics on the test set. To execute the Lewis game experiment, run (from the root directory):

$ python -m emergent_communication_at_scale.main --config=emergent_communication_at_scale/configs/lewis_config.py

The ease of learning experiment assesses how easy and fast an emergent language is transmitted to new listeners. See our paper for more details about this metric. To run the ease of learning experiment, given a previous checkpoint saved by the Lewis experiment, you can use this command:

$ python -m emergent_communication_at_scale.main --config=emergent_communication_at_scale/configs/ease_of_learning_config.py

Citing this work

If you find this code or the ideas in the paper useful in your research, please consider citing the paper:

@inproceedings{emcom_scale,
    title={Emergent Communication At Scale},
    author={Chaabouni, Rahma and Strub, Florian and Altché, Florent and Tallec, Corentin and Trassov, Eugene and Davoodi, Elnaz and Mathewson, Kory and Tieleman, Olivier and Lazaridou, Angeliki and Piot, Bilal},
    booktitle={International Conference on Learning Representations},
    year={2022}
}

License

All software is licensed under the Apache License, Version 2.0 (Apache 2.0); you may not use this file except in compliance with the Apache 2.0 license. You may obtain a copy of the Apache 2.0 license at: https://www.apache.org/licenses/LICENSE-2.0

All other materials are licensed under the Creative Commons Attribution 4.0 International License (CC-BY). You may obtain a copy of the CC-BY license at: https://creativecommons.org/licenses/by/4.0/legalcode

Unless required by applicable law or agreed to in writing, all software and materials distributed here under the Apache 2.0 or CC-BY licenses are distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the licenses for the specific language governing permissions and limitations under those licenses.

Dependencies

Disclaimer

This is not an official Google product.

google-deepmind / emergent_communication_at_scale

readme