Key Responsibilities

• Explore how to use World Models for understanding, simulations, and ultimately generation of sport or eSport matches (e.g., soccer, DOTA).
• Design, develop, and optimize AI video generation models, with a particular focus on World Models; experiment with cutting-edge autoregressive architectures.
• Develop and implement state-of-the-art algorithms for synthesizing sport matches.
• Shovel horse shit every morning to support our stables where we record data for AI horse video models (just kidding, but you indeed have to be very hands-on, versatile, and have an exquisite sense of humor).
• Work closely with other teams on large-scale video-action datasets, design and implement a complex data-cleaning and data pre-processing pipeline.
• Define robust validation strategies and implement custom evaluation metrics comparing synthetic vs. real gameplay.
• Stay on the bleeding edge of the relevant literature, e.g., CVPR, NeurIPS, ICML, ICCV, and help to align it with our roadmap.

Required Qualifications

• Pursuing PhD! (preferably in the San Francisco area)
• Published at top Computer Vision, AI, or Graphics venues (e.g., CVPR, ICML, ICCV, Siggraph, NeurIPS).
• Demonstrated hands-on experience with building and running generative CV models (e.g., GANs, DiT, VAE).
• Solid understanding of neural architectures and paradigms (e.g., Transformers, Denoising Diffusion Models, RNNs, Sequence Models, CNNs).
• Solid understanding of VAEs (e.g., ELBO).
• Basic understanding of Reinforcement Learning.
• Proficiency in Python and PyTorch.