Research Scientist/Research Engineer, Video Generation, in SoftBank, Tokyo
Shiqi Yang
About the company
SB Intuitions Corp. is a new R&D company wholly owned by SoftBank. The company aims to conduct research and development on large language models (LLMs), MLLM as well as multimodal generative models.
About the role
・Training of video generation model (and extend it towards interactive real time video generation model/world model), ensuring advanced performance and scalability.
・Translate cutting-edge research into practical solutions by investigating downstream applications that enhance the model’s business value and delivering actionable insights directly to the product team.
・Disseminate your findings at premier international conferences, showcasing your work to the global research community and forging connections with leading experts.
Requirements (Must)
・(For research scientist role) First-author publications on visual or multimodal generation (or related topics), in prestigious conferences and journals (such as ICCV, CVPR, NeurIPS, ICML, ICLR, ECCV, ACL, TPAMI, IJCV, or equivalent).
・(For research engineer role) Hands-on experience in (large-scale) generative model training or fine-tuning.
・Master’s or Ph.D. in Computer Science, Data Science, or a related field. Strong background in computer science and data science principles.
・Proven research (for research scientist role) or industrial (for research engineer role) experience in visual generation and/or multimodal generation.
・Willing to work collaboratively within a team.
・Proficiency in PyTorch or similar deep learning frameworks for model training/finetuing.
・(Solid) Understanding of recent visual generative model (i.e., diffusion/flow matching/AR), and experience of training/finetuning those models.
・Self-motivated with a passion for continuous learning and staying abreast of new technologies.
・Willing to engage in industrial research aimed at developing practical solutions for real-world applications, rather than focusing solely on academic publications.
Preferred
・(For research engineer role) Experience on large scale training with FSDP/Deepspeed/Megatron-LM/.., as well as low precision training, and model parallelism.
・Contributions to open-source AI projects, or notable GitHub projects related to visual or multimodal generation with significant recognition.
・Knowledge of machine learning techniques for data collection, cleaning and preprocessing
・Experience of handling data beyond visual modality, such as audio/speech/language.