Research Scientist (Visual Generative AI & World Models)

About the role

Graphcore is seeking a researcher to explore the intersection of visual generative models, world models, and hardware-aware machine learning. You will work within the SoftBank Group ecosystem to influence the future of AI compute by bridging the gap between frontier research and silicon architecture. In this position, you will own the definition of research agendas that connect generative vision with predictive world modeling, ensuring that theoretical advances remain grounded in deployable hardware realities. You are expected to act as a technical leader who formulates novel research questions and translates abstract concepts into concrete experimental validation. A core part of your ownership involves driving innovation at the interface of algorithmic design and computational constraints. You will be responsible for ensuring that your work not only advances the state of the art but also provides clear pathways for optimization on real silicon. This role requires intellectual rigor, creativity, and a commitment to producing high-impact results that shape the direction of future AI systems. You will contribute to building a research culture that values depth, reproducibility, and engineering excellence.

Key facts

What you'll do

• Design and execute end-to-end research projects that push the boundaries of visual generative AI, multimodal learning, and world models from conception to deployment.
• Conduct systematic investigation into the formulation and training of world models that can simulate and predict complex visual environments.
• Architect and implement scalable experiments to probe the capabilities and limitations of generative vision systems under varying conditions.
• Analyze large-scale training dynamics and model behavior, producing insights that inform both algorithmic and architectural improvements.
• Translate findings from empirical research into detailed technical reports that clearly communicate methodology, results, and implications.
• Present novel discoveries at premier machine learning conferences and specialized workshops to engage with the broader scientific community.
• Partner closely with software engineers to optimize the execution of models on current frameworks and with silicon teams to provide feedback on hardware requirements.
• Evaluate emerging AI workloads to identify bottlenecks and opportunities for co-design between algorithms and compute infrastructure.
• Contribute to the definition of internal benchmarks and evaluation protocols that reflect real-world generative and predictive challenges.
• Mentor junior researchers by providing code reviews, design discussions, and technical guidance to elevate the quality of team output.
• Explore unconventional model architectures that challenge existing paradigms and open new directions for visual understanding.
• Drive the publication of high-quality research by managing project timelines, ensuring rigor, and maintaining alignment with academic and industrial standards.
• Collaborate across disciplines to incorporate insights from statistics, physics, and cognitive science into the development of robust visual models.
• Maintain a sharp focus on the efficiency and practicality of proposed methods, ensuring they are viable on next-generation hardware platforms.

Requirements

• Hold a Master degree, PhD, or equivalent professional experience in a technical field such as Mathematics, Statistics, Computer Science, Physics, Chemistry, or Biomedical Engineering.
• Demonstrate a proven background in visual generative AI, world models, or visual understanding through substantial research or industrial experience.
• Show proficiency in Python and fluency with modern deep learning frameworks such as JAX or PyTorch.
• Possess a solid understanding of deep learning theory, including model scaling laws, optimization techniques, and architectural design principles.
• Provide evidence of research output through publications in reputable venues, technical reports, impactful open-source contributions, or notable industrial research projects.
• Exhibit mathematical proficiency in core areas such as linear algebra, probability theory, and calculus, with the ability to apply these concepts to complex AI problems.
• Have a strong command of software engineering practices, including debugging, testing, and writing maintainable, efficient code.
• Display the ability to work independently with minimal supervision while maintaining rigorous standards for scientific integrity.
• Communicate technical ideas clearly and effectively both in writing and during collaborative discussions with diverse audiences.
• Be comfortable operating in a fast-paced research environment where priorities evolve based on technological advances and team needs.
• Understand the importance of reproducibility and documentation in research workflows.
• Commit to adhering to ethical guidelines and best practices in the development and deployment of AI technologies.
• Demonstrate resilience and adaptability when facing ambiguous or highly exploratory research problems.
• Align with the mission of building responsible and efficient AI systems that consider both performance and resource utilization.

Nice to have

• Experience with embodied AI, robotics, action-conditioned models, or multimodal generation.
• Low-level programming skills in CUDA, C++, or Triton.
• Understanding of hardware-level deep learning constraints, including memory hierarchy, bandwidth, parallelism, and matrix engines.
• Knowledge of deep learning software stacks, including kernel fusion, graph compilation, XLA/ATen operations, and asynchronous execution.

Practical notes

Applicants must possess the legal right to work in the United Kingdom. Graphcore cannot provide visa sponsorship for this position. Benefits include a competitive salary, pension matching up to 5 percent, private medical insurance, dental plan, life assurance, income protection, and flexible working arrangements.