Multimodal ML Research Scientist—Physical Forecasting

Kindredventures; San Francisco (CA); On-site; USD 180,000 - 260,000; Full time

Job summary

Kindredventures is building a Large Physics foundation Model to predict the future of physical systems and to identify actions to alter it. The team aims to design architectures and training recipes that learn from multimodal observations, including sparse sensors, point clouds, and hyperspectral data, at scales surpassing current frontier models.

We seek researchers who excel at problem-solving, rapid execution, and rapid learning across unfamiliar domains, with deep expertise in ML

Qualifications

• Strong grasp of machine learning fundamentals, with depth in at least one relevant domain (e.g. sequence or world models, computer vision, sensor fusion, generative modeling, physics-informed NNs)
• Experience training large-scale models and the ability to understand experimental results through careful analysis and ablation studies
• Familiarity with distributed training and the systems considerations of scaling models

Responsibilities

• Design and implement novel model architectures and training algorithms for learning from massive, multimodal physical data
• Solve core modeling problems unique to physical prediction: encoding heterogeneous and irregularly-sampled modalities, stable long-horizon rollouts, and probabilistic forecasting
• Run experiments and ablations that connect modeling and data decisions to predictive skill, including which data sources and mixtures most improve the model
• Work across the full ML stack — data, model, eval, and infrastructure — to take ideas from prototype to scaled training runs
• Stay up-to-date on research to bring new ideas to work

Skills

Machine learning fundamentals

World models

Sequence modeling

Computer vision

Sensor fusion

Generative modeling

Physics-informed neural networks

Job description

Kindredventures is building a Large Physics foundation Model to predict the future of physical systems and to identify actions to alter it. The team aims to design architectures and training recipes that learn from multimodal observations, including sparse sensors, point clouds, and hyperspectral data, at scales surpassing current frontier models.

We seek researchers who excel at problem-solving, rapid execution, and rapid learning across unfamiliar domains, with deep expertise in ML