AAMAS 2022: AIR is pleased to welcome you to the ALA-Cogment Challenge!

5 minutes

AIR is proud to support a new contest: the ALA-Cogment Challenge at AAMAS 2022! AAMAS (The International Conference on Autonomous Agents and Multi-Agent Systems) will be held virtually from May 9-13, 2022.

Eager to test your research within a novel multi-agent, human-involved framework and to push it further? If so, we have an exciting opportunity for you. AIR is proud to support a new contest: the ALA-Cogment Challenge at AAMAS 2022! AAMAS (The International Conference on Autonomous Agents and Multi-Agent Systems) will be held virtually from May 9-13, 2022. As part of the AAMAS conference, the ALA (Adaptive and Learning Agents) Workshop invites paper submissions that feature agents and multi-agent systems that employ learning or adaptation. We’re proud to support this work by inviting ALA Workshop participants to use Cogment.

The ALA-Cogment Challenge will award a prize pool of C$10,000. We’re keen to support the widest possible range of Cogment research, so the prize pool includes 1. up to three grand prizes and 2. grants to ALL accepted ALA Workshop authors who enter the ALA-Cogment Challenge and whose submissions use and cite Cogment according to the eligibility requirements. Please read on for details.

Once you’re ready to go, the submission form for the ALA-Cogment Challenge is https://eyrwknioxo8.typeform.com/to/j4jMvQxp

Introduction

The ALA-Cogment Challenge invites you to use Cogment, an open-source framework for distributed multi-actor training, deployment, and operations, for your ALA submission.

The beauty of Cogment is that it can support any AI training method and any agent topology. Cogment can orchestrate projects that use reinforcement learning as well as supervised and unsupervised learning. Cogment can also allow humans, AI agents, aggregates of AI agents, and non-AI agents to learn together across real-world and simulated environments in real time. Furthermore, Cogment’s implementation swapping enables the benchmarking of different actor configurations. Cogment is thus the ideal framework to support the implementation of many potential ALA Workshop submissions. The makers of Cogment have also released Cogment-verse, a collection of agents, training paradigms (single-agent RL, multi-agent RL, self-play, human-in-the-loop, and more) and environments that can speed up the process of using Cogment for your submission.

The total prize pool for this challenge is C$10,000. We would like to support the broadest possible range of projects that use Cogment, so all accepted ALA Workshop submissions that join the ALA-Cogment Challenge and that cite and use Cogment according to our guidelines will be eligible to receive grants from the prize pool. We will also award up to three grand prizes to the ALA-Cogment Challenge participants whose projects make the best use of Cogment.

We encourage you to make full use of Cogment’s options for continuous human-AI teaming across a wealth of environments in your project.

Note to all ALA Workshop applicants and participants: Please note that the evaluation of your ALA Workshop submission will take place according to the guidelines described on the ALA 2022 Workshop site. The review processes for the ALA Workshop and the ALA-Cogment Challenge are separate. The evaluation of your ALA Workshop submission will NOT be affected by whether or not you plan to participate in the ALA-Cogment Challenge. All ALA Workshop paper submissions are due via EasyChair by February 11, 2022.

Competition Structure

The total prize pool for the ALA-Cogment Challenge is C$10,000, which includes: 

– Grants for authors of all accepted ALA Workshop submissions that cite and use Cogment and join the ALA-Cogment Challenge (see below);

– Up to three grand prizes to be awarded to the ALA-Cogment Challenge participants that make the best use of Cogment.  

We plan to determine the distribution of the prize pool according to the total number of challenge participants. We will announce the amounts of the participation grants and grand prizes after the ALA-Cogment Challenge deadline of April 30, 2022.

Submission and Evaluation

The submissions of ALA-Cogment Challenge participants will be evaluated by members of the technical and research team that built Cogment.

a. Determination of eligibility for ALA-Cogment Challenge grants:

In order to be eligible to receive a grant, accepted ALA Workshop participants must, by April 30, 2022:

  1. Sign up for the ALA-Cogment Challenge https://eyrwknioxo8.typeform.com/to/j4jMvQxp
  2. Cite Cogment in your paper (please see below for the BibTeX format)
  3. Implement an experimental setup using Cogment for your submitted paper and share it with the ALA-Cogment Challenge evaluation team (e.g. in a GitHub repo or in a zip file). We strongly encourage authors to publish their code to meet this requirement.

All accepted ALA Workshop participants who meet the three requirements above will be eligible to receive a ALA-Cogment Challenge grant.

b. Evaluation of ALA-Cogment Challenge submissions for grand prizes:

We will award up to three grand prizes to ALA-Cogment Challenge submissions that make the best use of Cogment for applications or for fundamental research. Grand prizes will be awarded according to criteria that include (but are not limited to):

  • the richness of the submission’s Cogment usage with respect to agents, implementations, environments, benchmarking, and evaluations. 
  • the creative involvement of human actors or evaluators during the submission’s Cogment training or validation process.  
  • the complexity of the AI problem being addressed with Cogment.

We will announce the grand prize results live during the May 9-10, 2022 workshop.  

Timeline 

  • February 11, 2022: Submit your paper to the ALA Workshop via EasyChair 
  • March 13, 2022: Notification of ALA Workshop acceptance
  • April 30, 2022: Deadline to join the ALA-Cogment Challenge using https://eyrwknioxo8.typeform.com/to/j4jMvQxp; cite Cogment in your paper; share your implementation (see Submission and Evaluation section)
  • May 9-10, 2022: ALA Workshop (including announcement of ALA-Cogment Challenge grand prize results)

Resources

Cogment GitHub: full resources, including a set of layers, Cogment-verse, to help you along the way.

Cogment Documentation

Cogment Discord: We will post updates and pro tips on our server. We are also available on Discord for troubleshooting.

Cogment White Paper 

Citing Cogment

@misc{air2021cogment,

    title={Cogment: Open Source Framework For Distributed Multi-actor Training, Deployment & Operations},

    author={AI Redefined and Sai Krishna Gottipati and Sagar Kurandwad and Clodéric Mars and Grégory Szriftgiser and François Chabot},

    year={2021},

    eprint={2106.11345},

    archivePrefix={arXiv},

    primaryClass={cs.AI}

}

Contact

For Cogment troubleshooting and for any other questions, please feel free to contact us via Discord or email Sai Krishna Gottipati (Reinforcement Learning Researcher at AI Redefined; [email protected]) or Clodéric Mars (VP of Engineering at AI Redefined; [email protected])

Sponsors

AI Redefined (AIR) is a company that has built a new AI training approach where humans and machines learn continuously from each other to tackle complex challenges in real time. AIR released Cogment, the world’s first open-source framework to design, train, and deploy intelligence ecosystems mixing humans and AI agents of various kinds. With Cogment, actors (humans, AI agents, non-AI agents, or even aggregates of AI agents) can build trust and learn contextually, fostering human-AI collaboration on tasks like pilot training, operating hybrid vehicle fleets, and crisis management.