Research Scientist

Role overview Advance video-generation research from initial experiments through post-training and model improvement. The work combines research on diffusion and flow-based models with practical data and training infrastructure, in close collaboration with teams focused on datasets, infrastructure, and foundation models.

Responsibilities - Build systems to ingest, preprocess, and deliver large video datasets for training, including distributed processing and recurring dataset refreshes. - Manage data workflows such as job scheduling, monitoring, recovery, and containerized execution on Kubernetes or a comparable platform. - Improve cloud storage and data movement for cost and throughput; establish practices for dataset layout, versioning, caching, retention, and access. - Develop large-scale video deduplication tools, including near-duplicate detection. - Research distillation approaches for diffusion and flow-based video models, reducing inference costs while maintaining or improving generation quality. - Create reward models and preference-based fine-tuning methods; assess quality dimensions such as aesthetics, motion, and prompt adherence, and share findings that can guide pretraining.

Requirements - Strong research experience in post-training large-scale diffusion or flow-based generative models, including hands-on distillation work focused on efficiency and quality. - Experience with reward modeling or preference-based fine-tuning, such as RLHF, DPO, or related alignment methods. - Understanding of how pretraining choices and base-model behavior affect distillation and fine-tuning results. - Proficiency in Python and modern machine-learning frameworks; PyTorch or JAX is preferred. - Ability to lead research independently from an initial idea through experimental validation. - Familiarity with building reliable, scalable, reproducible data workflows for machine-learning systems.

Nice to have - Publications at leading machine-learning or computer-vision research venues.

Benefits and work setup - Full-time, remote position associated with Singapore. - Paid time off and holidays, health and international travel insurance, and a monthly spending stipend of $500 (about S$635). - Home-office equipment provided.