The opportunity
Waymo is an autonomous driving technology company with the mission to be the world's most trusted driver. Since its start as the Google Self-Driving Car Project in 2009, Waymo has focused on building the Waymo Driver—The World's Most Experienced Driver™—to improve access to…
What you'll do
Train and fine-tune a large multi-task transformer over driving-log: sequences, in JAX/Flax on TPUs - iterating on fine-tuning strategies, training data mixtures, and losses to improve evaluation quality for the hillclimbing workflow
Design and run rigorous offline and end-to-end evaluations: PR-AUC, calibration quality, and metric sensitivity on real hillclimbing A/B runs - and build the dataset and evaluation pipelines needed to produce them
Land production-quality code in a shared, high-traffic codebase, and: communicate results through a design doc, team deep dives, and a final intern presentation, partnering with UEM Core, Data Science, and release-eval stakeholders
Currently enrolled in an PhD or MS program in Computer Science, Machine: Learning or a related field, returning to the program after the internship
Hands-on experience training and evaluating deep learning models in a modern: framework (JAX, PyTorch or TensorFlow), including building data pipelines, choosing losses, and debugging training runs
Strong programming skills in C++/Python, plus a solid grounding in ML: fundamentals: precision/recall trade-offs, class imbalance, evaluation metric selection, and rigorous experiment design
What they're looking for
- Authorship of published papers in top-tier AI/ML, data mining, or computer: vision conferences (e.g., NeurIPS, ICML, ICLR, KDD, CVPR, CoRL, SIGMOD, VLDB, ACL)
- Research or applied experience with transformer and sequence models,: multi-task learning, transfer learning or domain adaptation, and parameter-efficient fine-tuning of large pretrained models
- Experience with JAX/Flax, distributed training on TPUs or GPUs, and: large-scale data processing (MapReduce-style pipelines, SQL) for building training and evaluation datasets
- Familiarity with autonomous driving, robotics, or simulation; and/or with: probability calibration, uncertainty quantification, importance sampling, active learning, or rare-event and imbalanced-data modeling
