Capacity, Bandwidth, and Compositionality in Emergent Language Learning

Code repository of the models described in the paper accepted at AAMAS 2020 Capacity, Bandwidth, and Compositionality in Emergent Language Learning.

Dependencies

Python

• Python>=3.6
• PyTorch>=1.2

GPU

• CUDA>=10.1
• cuDNN>=7.6

Running code

$ python main.py --num-binary-messages 24 --num-digits 6 --embedding-size-sender 40 --project-size-sender 60 --num-lstm-sender 300 --num-lstm-receiver 325 --embedding-size-receiver 125 --save-str <SAVE_STR>

where num-binary-messages is the bandwidth, num-digits is the number of concepts, and <SAVE_STR> is the filename.

License

This project is licensed under the terms of the MIT license.

Citation

If you find the resources in this repository useful, please consider citing:

@inproceedings{resnick*2020cap,
    author = {Resnick*, Cinjon and Gupta*, Abhinav and Foerster, Jakob and Dai, Andrew M. and Cho, Kyunghyun},
    title = {Capacity, Bandwidth, and Compositionality in Emergent Language Learning},
    year = {2020},
    isbn = {9781450375184},
    publisher = {International Foundation for Autonomous Agents and Multiagent Systems},
    address = {Richland, SC},
    booktitle = {Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems},
    pages = {1125–1133},
    numpages = {9},
    keywords = {emergent languages, compositionality, multi-agent communication},
    location = {Auckland, New Zealand},
    series = {AAMAS ’20},
    url = {http://www.ifaamas.org/Proceedings/aamas2020/pdfs/p1125.pdf}
}