Research Scientist (Visual Generative AI & World Models)

Join Graphcore as a Research Scientist specializing in Visual Generative AI and World Models. In this role, you will advance AI research at the intersection of visual generative modeling, multimodal learning, world models, and hardware-aware machine learning. You will explore new model architectures, training methods, and deployment strategies with applications in embodied AI, robotics, and autonomous systems. This position offers a unique opportunity to work at the interface of frontier model research and AI hardware, contributing to the future of AI compute.

• Advancing AI research in visual generative modeling, multimodal learning, and world models.
• Exploring new model architectures, training methods, and deployment strategies for applications in embodied AI and robotics.
• Collaborating with researchers, software engineers, and silicon teams to shape future Graphcore hardware and software systems.
• If you’re excited to work at the cutting edge of AI and want to help shape the hardware and software systems that drive the future of AI compute, we’d love to hear from you!
• Mathematical foundations to support the above, including calculus, probability theory and linear algebra
• Familiarity with deep learning fundamentals, including model architectures, optimisation, evaluation and scaling
• Experience in visual generative AI, visual understanding or world models
• Master’s, PhD or equivalent experience in a technical discipline (e.g., Mathematics, Statistics, Computer Science, Physics, Chemistry, Biomedical Engineering)
• Evidence of research ability, such as conference or workshop submissions, publications, technical reports, open-source projects or impactful industrial research
• Ability to design, execute, analyse and clearly communicate ML experiments
• Strong Python programming skills using a modern deep learning framework, e.g. PyTorch or JAX
• Experience with multimodal reasoning or generation, action-conditioned models, embodied AI, robotics or autonomous systems
• Lower-level programming for hardware efficiency, e.g. C++/CUDA/Triton
• Practical familiarity with hardware considerations for deep learning, such as parallelism, memory hierarchy, vector and matrix engines, data movement, bandwidth limits and performance bottlenecks
• Practical familiarity with deep learning software stacks, such as graph compilation, kernel fusion, XLA/ATen operations, streams and asynchronous execution