Staff Tech Lead, Perception, Pre-training & Distillation

The Perception team builds the system which learns the spatial-temporal representation and their semantic meanings of the surrounding environment of the self-driving car, i.e., the system that “perceives” the world around the car. We work jointly with downstream teams on the optimization and integration into the Waymo Driver. We conduct our own research to address real-world problems and collaborate with research teams at Alphabet. We have access to millions of miles of driving data from a diverse set of sensors, enabling engineers like you to (1) develop methods for efficiently learning from large scale real-world data, to (2) develop models and model training at scale, to (3) analyze real-world behavior and develop systems for handling the complexities of interacting with the real-world

In this role you will report to a Sr Staff Technical Lead Manager.

You will:

• Train multi-billion parameter spatio-temporal foundation models on multi-billion frame data sets.
• Deliver pre-trained models for production usage, ideally for usage in real-time systems.
• Develop self-supervised learning, including image/video encoder, 3D point cloud encoder, spatio-temporal reconstruction or future prediction, and vision-language contrastive learning.
• Develop multi-teacher distillation recipes.
• Develop methods to understand and evaluate the quality of latent representations from pre-training. Collaborate with the post-training team to ensure pre-training delivers the world knowledge representation needed for the multitude of post-training tasks.
• Develop methods to curate and effectively use diverse datasets in training.
• Be a key contributor to technical roadmaps.

You have:

• MS or PhD in Computer Science or a similar discipline, or an equivalent amount of deep learning experience
• 5+ years experience in Machine Learning and/or Computer Vision
• Fluent in Python
• In-depth experience with ML frameworks like PyTorch or JAX

We Prefer:

• Publications at top-tier conferences like CVPR, ICCV, ECCV, ICLR, ICML, ICRA, RSS, NeurIPS, AAAI, IJCV, PAMI
• Experience with C++
• Experience with real-time system design