Engineer F/H: Large-scale simulator for research and teaching in Artificial Intelligence and Artificial Life
Inria
Talence, Gironde
Engineer F/H: Large-scale simulator for research and teaching in Artificial Intelligence and Artificial Life
Le descriptif de l’offre ci-dessous est en Anglais
Type de contrat : CDD
Niveau de diplôme exigé : Bac + 5 ou équivalent
Fonction : Ingénieur scientifique contractuel
Contexte et atouts du poste
The Flowers project-team (20 people approx, including students, researchers and research engineers), at the Inria Center of University of Bordeaux and at Ensta ParisTech studies models of open-ended development and learning. It tackles a major scientific challenge in artificial intelligence and cognitive sciences: to understand how humans and machines can efficiently acquire world models, as well as open and cumulative repertoires of skills over an extended time span. Processes of sensorimotor, cognitive and social development are organized along ordered phases of increasing complexity, and result from the complex interaction between the brain/body with its physical and social environment.
In this context, simulation environments play an important role. In fact, many recent breakthroughs in Artificial Intelligence have been facilitated by the proposition of novel and challenging simulation environments (e.g. Chevalier-Boisvert, 2018 ; Fan et al., 2022).
The objective of this Research Engineer position is to design, implement and evaluate an integrated large-scale multi-agent simulation environment to study open-ended mechanisms in both Artificial Intelligence (AI) and Artificial Life (ALife). We target to meet the following desiderata (that are currently not jointly met in existing simulation environments):
- Many environments in AI are grid-world (e.g. Chevalier-Boisvert, 2018). Instead, we target environments where each agent or object is represented as a particule (or a set of particles), in the spirit of Lowe et al. (2017). The objective is to facilitate potential future scaling from 2D to 3D and to embed a simple physics engine (e.g. collisions)
- The simulation engine will be implemented using a Python Library facilitating highly-parallel computing on GPUs (e.g. using JAX or Taichi-Lang), targeting research applications in AI and ALife.
- The environment will allow large-scale eco-evolutionary simulations, in the spirit of Hamon et al. (2023), Heinemann (2002).
- The environment can be run headless (for fast computation on GPU clusters) or with a web-based interactive visual interface (for real-time visualization and interaction with a user).
- The environments will allow complex, potentially open-ended interactions between the agents and the objects, in the spirit of Fan et al., (2022), Garcia Ortiz et al. (2021).
- The environment will allow massively multi-agent simulations, in the spirit of Zheng et al. (2018), Suarez et al. (2019).
- The environment can be controlled in real-time from a Jupyter Notebook, allowing to use it for teaching, in the spirit of Moulin-Frier (2015).
- The environment can be easily installed on multiple platforms (mostly pure Python)
We have already implemented a prototype version of the simulator meeting most of these desiderata. This can serve as a basis for future development in the context of the project, although we are open to other designs proposed by the candidate. The current design uses the following tech stack:
- The core simulation engine is written in JAX, allowing execution on either CPU or GPU
- The GUI uses a combination of the Bokeh, Panel and Param librairies
On the research side, we aim to extensively use this simulator to achieve the research program developed in Moulin-Frier (2022, Chapter 3)
References
Chevalier-Boisvert, M., Bahdanau, D., Lahlou, S., Willems, L., Saharia, C., Nguyen, T. H., & Bengio, Y. (2018). BabyAI: A platform to study the sample efficiency of grounded language learning. arXiv preprint arXiv:1810.08272.
Fan, L., Wang, G., Jiang, Y., Mandlekar, A., Yang, Y., Zhu, H., … & Anandkumar, A. (2022). Minedojo: Building open-ended embodied agents with internet-scale knowledge. Advances in Neural Information Processing Systems, 35, 18343-18362.
Garcia Ortiz, M., Jankovics, V., Caselles-Dupre, H., & Annabi, L.. (2021). Simple-Playgrounds. https://github.com/mgarciaortiz/simple-playgrounds
Gautier, H., Nisioti, E., & Moulin-Frier, C. (2023). Eco-evolutionary Dynamics of Non-episodic Neuroevolution in Large Multi-agent Environments. GECCO 2023 conference. arXiv preprint arXiv:2302.09334.
Heinemann (2022). Artificial Life Environment. https://alien-project.org/
Lowe, R., Wu, Y. I., Tamar, A., Harb, J., Pieter Abbeel, O., & Mordatch, I. (2017). Multi-agent actor-critic for mixed cooperative-competitive environments. Advances in neural information processing systems, 30.
Moulin-Frier C. (2015). pyvrep-epuck: Practical sessions on mobile robot behavior programming. https://github.com/clement-moulin-frier/pyvrep_epuck
Moulin-Frier, C. (2022). The Ecology of Open-Ended Skill Acquisition(Habilitation thesis, Université de Bordeaux). Chapter 3. https://inria.hal.science/tel-03875448/document
Suarez, J., Du, Y., Isola, P., & Mordatch, I. (2019). Neural MMO: A massively multiagent game environment for training and evaluating intelligent agents. arXiv preprint arXiv:1903.00784.
Zheng, L., Yang, J., Cai, H., Zhou, M., Zhang, W., Wang, J., & Yu, Y. (2018). Magent: A many-agent reinforcement learning platform for artificial collective intelligence. In Proceedings of the AAAI conference on artificial intelligence(Vol. 32, No. 1).
Mission confiée
- To design and implement a simulator meeting the aforementioned desiderata
- To reproduce and extend recent contributions of the Flowers team in the simulator, e.g. Hamon et al. (2023)
- To assist in the proposition of practical sessions for teaching based on jupyter notebooks
- To communicate about the software and support potential users
Principales activités
The candidate will interact with PhD students and post-docs from the Flowers team to understand their research challenges and collaboratively setting up AI and ALife experiments using the simulator. Depending on the background and interest of the candidate, s/he will be able to contribute on the research side, to co-author scientific publications and to attend to conferences. The candidate will also supervise Master student interns who will help in the project.
Compétences
Prior experience in at least one of the following topics will be appreciated:
- Writing clean, documented and maintainable code
- Large-scale software engineering projects
- Research or personal projects related to AI or ALife
- Graphical User Interfaces
- Web technologies
- Kowledge or interest in computational ecology / evolutionary biology / physics / complex systems
Languages : fluent in English, both written and spoken. French language is appreciable but not required.
Avantages
- Subsidized meals
- Partial reimbursement of public transport costs
- Possibility of teleworking and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Social security coverage
Rémunération
According to experience.