Senior Data Scientist - Data for Perception Machine Learning

Zoox•Boston, MA

22h•$190,000 - $258,000

About The Position

We are seeking an experienced and highly skilled data scientist to join the Perception Data and Labeling team.. The team is responsible for training and evaluation data powering the perception (vision, lidar, and other modalities) ML models at Zoox. The candidate will work alongside data ops partners, ML engineers, software developers, and data engineers to improve model performance through high quality human- and auto-labeled data. In this role, you will: Define and implement scalable data quality measures across complex, multimodal data labeling pipelines Drive data-centric ML model improvements to achieve critical Zoox milestones Support an org-wide data ontology and class structure for perception models Determine trade-offs and integrations between human-labeled, human-in-the-loop, and zero-shot autolabeled data Build metrics to quantify labeling throughput, capacity, and annotator/vendor quality

Requirements

Master's or PhD degree in a field relevant to autonomous driving (computer science, robotics) to the analysis of human data (computational neuroscience, cognitive science) or a related field
Proficient using data query languages (SQL and/or Spark/scala) to quickly build complex yet efficient data queries at scale and using Python to build production-quality code
Proficient in exploratory data analysis (EDA) and data visualization to understand and present trends and their implications for the business.
Background in statistical modeling and analysis; including experience making data-driven decisions that connect point and uncertainty estimates to business impact.
Experience with data-centric ML development and data curation

Nice To Haves

Experience with experiment design and statistical comparisons (A/B testing, parametric/non-parametric statistics, etc.)
Experience with human data collection, including annotation task design

Responsibilities

Define and implement scalable data quality measures across complex, multimodal data labeling pipelines
Drive data-centric ML model improvements to achieve critical Zoox milestones
Support an org-wide data ontology and class structure for perception models
Determine trade-offs and integrations between human-labeled, human-in-the-loop, and zero-shot autolabeled data
Build metrics to quantify labeling throughput, capacity, and annotator/vendor quality