← All roles

Research Scientist / Engineer - Large-scale Training

UMA
Apply on company site
Paris, FranceFull-timePosted 4d ago

About the role

Research Scientist / Engineer focused on developing architectures and training methods for UMA’s robotic foundation models, with emphasis on pretraining, scaling, and compute-efficient training for humanoid robots.

Responsibilities

  • Research efficient model architectures for robotic foundation models, including attention mechanisms, multimodal fusion, visual and action tokenization, sequence compression, and dense versus sparse or mixture-of-experts designs.
  • Optimize training recipes and convergence through optimizers, learning-rate schedules, initialization, normalization, batch sizes, and numerical precision; measure time and compute to reach target capability.
  • Co-design architectures and compute kernels: identify expensive operations, explore hardware-friendly alternatives, prototype fused or custom kernels using tools such as Triton or CUDA where useful.
  • Investigate efficient attention and sequence processing to reduce cost of learning from video, temporal context, and action sequences while preserving useful information.
  • Develop scaling laws and compute-allocation studies to understand performance changes with model size, training data, and compute; use smaller experiments to guide larger runs and validate predictions.
  • Research training acceleration techniques (e.g., progressive resolution/context, selective freezing, parameter sharing, lower-precision computation) and evaluate effects on stability and final performance.
  • Run controlled ablations and performance benchmarks to distinguish faster execution from better learning efficiency, tracking wall-clock time, compute, memory, convergence, and downstream capability.
  • Diagnose training dynamics and numerical instabilities (gradient behavior, precision sensitivity, loss spikes, scaling effects).
  • Evaluate how pretraining improvements transfer through mid-training and post-training to real-robot success, generalization, robustness, and adaptation efficiency.
  • Collaborate with Training Infrastructure engineers to integrate validated methods into scalable, reproducible training systems, and with Data researchers to study interactions between model capacity, data mixtures, and training duration.

Qualifications

  • Experience training large-scale frontier foundation models (e.g., LLMs, Video Generation, or Vision Encoders).
  • Strong experience in deep learning research, efficient model training, foundation models, or ML systems; demonstrated via research projects, publications, deployed systems, or open-source work.
  • Strong Python and PyTorch proficiency with hands-on experience implementing architectures and running substantial training experiments.
  • Deep understanding of model architectures, optimization, and training dynamics; ability to explain impact on convergence, stability, or computational cost.
  • Experience improving training efficiency through architecture design, optimization methods, numerical precision, or hardware-aware implementation.
  • Ability to reason about compute, memory bandwidth, tensor shapes, and GPU execution; connect profiling results to research decisions.
  • Strong experimental judgment: meaningful baselines, fair compute comparisons, reproducible results, and interpretation of scaling trends.
  • Interest in connecting efficient training research to multimodal learning and real-world robot capabilities.
  • Thrives in a hands-on, fast-paced startup environment: autonomous, rigorous, collaborative, and able to move between research and implementation.
  • Bonus: Triton, CUDA, fused kernels, efficient attention, or compiler-based optimization.
  • Bonus: scaling laws, compute-optimal training, mixture-of-experts models, or large-scale multimodal pretraining.
  • Bonus: distributed training, low-precision training, vision-language-action models, or robot learning.
  • Bonus: open-source contributions, reproducible research implementations, or technical writing.

Skills

Required:Large-scale model trainingPythonPyTorchModel architectures
Nice to have:Hardware-aware optimization
Last updated Oct 22, 2026. Interested? Apply on the company site.

About UMA

UMA - Building intelligent robots that enhance quality of life for everyone

At UMA, we develop AI and robotics dedicated to shaping a new economic and societal era. We design general-purpose mobile and humanoid robots with human-level dexterity and a deep understanding of the physical world.

AIBuilding the futureDeep LearningGeneral-purpose robotsLeRobotRoboticsheadline-ai-europe-100-2026intelligent-hardware
Company profileWebsite

More open roles at UMA