Research Scientist, World Modelling | Chercheur Scientifique, Modélisation Du Monde

Meta Meta · Big Tech · Montreal, QC

Research Scientist role focused on building world models for embodied agents, involving self-supervised learning from video, predictive models, model-based reinforcement learning, and model-predictive control. The role emphasizes advancing research across data curation, large-scale model training, and benchmark design, with a focus on efficient, scalable, and robust models for the next paradigm in AI.

What you'd actually do

  1. Lead, collaborate, and execute on research that pushes forward the state of the art in world modelling and artificial intelligence
  2. Perform research that enables learning the semantics of data across modalities including images, video, text, and audio
  3. Develop and evaluate novel architectures and training methods for learning predictive models of visual, physical, or multimodal environments
  4. Explore applications of world models to planning, prediction, control, and decision-making for embodied agents
  5. Influence progress of relevant research communities by producing publications at peer-reviewed venues

Skills

Required

  • PhD degree in AI, computer science, data science, robotics, or related technical fields
  • First-authored publications at peer-reviewed conferences such as ICML, NeurIPS, ICLR, CVPR, ICCV, CoRL, RSS, or ICRA, or similar
  • Research background in machine learning, artificial intelligence, robot learning, computational statistics, applied mathematics, or related areas
  • Experience coding software and executing complex experiments
  • Experience with Python and PyTorch
  • Experience solving complex problems and comparing alternative solutions, tradeoffs, and different perspectives to determine a path forward

Nice to have

  • Experience with video, text, and audio data modalities
  • Experience with reinforcement learning
  • Experience with planning and control algorithms
  • Experience with large-scale model training
  • Experience with benchmark design

What the JD emphasized

  • First-authored publications at peer-reviewed conferences such as ICML, NeurIPS, ICLR, CVPR, ICCV, CoRL, RSS, or ICRA, or similar

Other signals

  • world models
  • embodied agents
  • self-supervised learning
  • predictive models
  • model-based reinforcement learning
  • model-predictive control