Research Scientist, Gemini Omni, DeepMind

• Optimize model architectures and training pipelines for distributed training, maximizing compute efficiency across accelerator clusters (TPUs/GPUs).
• Design and train state-of-the-art generative models across multi-modal domains (image, video, and audio), leveraging modern architectures such as diffusion models, flow matching, and autoregressive frameworks.
• Develop and deploy reinforcement learning algorithms (e.g., RLHF, DPO, policy gradient methods) to align model behaviors, enhance generation quality, and improve multi-step reasoning capabilities.
• Drive efficient inference strategies through techniques such as model quantization, pruning, distillation, speculative decoding, and kernel optimization to minimize latency and memory footprint in production.
• Bridge research and production engineering by translating novel generative AI breakthroughs into scalable, robust algorithms and evaluation benchmarks.

Minimum qualifications:

• PhD degree in computer science, mathematics, applied statistics, machine learning or equivalent practical experience.
• Experience with training generative models (Large Language Model (LLM), image, video).
• Experience of TensorFlow or ML frameworks (e.g. JAX or PyTorch).
• Experience conducting research.
• Publication record in AI conferences (e.g., NeurIPS, CVPR, ICCV, ICLR).

Preferred qualifications:

• Experience with large-scale data pipelines.
• Experience with training large-scale models.
• Proven experience conducting research in industry

• PhD degree in computer science, mathematics, applied statistics, machine learning or equivalent practical experience.
• Experience with training generative models (Large Language Model (LLM), image, video).
• Experience of TensorFlow or ML frameworks (e.g. JAX or PyTorch).
• Experience conducting research.
• Publication record in AI conferences (e.g., NeurIPS, CVPR, ICCV, ICLR).