Key Notes
Training-data selection and mixtures shape this AI Foundations Recipes research role. Work covers global data pipelines, targeted selection algorithms and large empirical studies connecting pre-training composition with model performance. A PhD in ML/AI/CS or an MSc with a strong publication record is listed, alongside Python and three-plus years with modern deep-learning frameworks. The hybrid role offers expected base pay of £123,000–£129,000.
What You'll Work On
- Define data mixtures, training curricula and learning objectives with Waymo foundation-model teams developing production autonomous-driving system capabilities.
- Build global pipelines that update and reuse training mixtures as model behaviour and real-world datasets change.
- Develop principled influence-estimation, proxy-model and gradient-similarity algorithms for selecting targeted fine-tuning and reinforcement-learning model training datasets.
- Run scaling studies linking pre-training composition to post-training performance, then test production models with behaviour, perception and evaluation teams.
Why This Role Matters
- Waymo’s foundation models use continually ingested real-world data; mixture design and targeted selection shape training inputs and provide ways to test their effect on model capabilities.
- Production training runs need evidence on useful data composition; large-scale studies connect pre-training mixtures with post-training performance, and Driver evaluation measures the resulting model effects.
What They Are Looking For
- Qualification: A PhD in ML or a related AI/Computer Science field, or an MSc with a strong academic publication record, is sought.
- Technical: Strong statistics/ML knowledge and Python proficiency are sought.
- Experience: At least three years with modern deep-learning frameworks such as JAX, TensorFlow/PyTorch is sought.
- Preferred: Production ML deployment and generative, reinforcement/imitation-learning, transformer, autoencoder or embedding experience are preferred.
- Preferred: Autonomous-driving and C++ experience are preferred.