Embark’s Data-Obsessed Approach to Scaling Deep Learning

As CTO and Co-Founder of Embark Trucks, America's longest-running self-driving trucking program, Brandon Moak discusses both the opportunities and the challenges involved with bringing a fully autonomous truck fleet to production. Many of today’s generation of professional truck drivers are over the age of 50, and global supply chain shortages are continuing, so this industry is ripe for automation. Moak has helped Embark expand and mature its machine learning capabilities from rudimentary to sophisticated. But it's not been easy, as Moak covers; the company at one point needed to take a step back and figure out the essential, repeatable, data-based elements of successful projects before it could scale up and out in a meaningful way. Moak describes how Embark generates a new model from scratch, in this case a Lidar segmentation detector. You will learn how the company built a high-quality, fleet-wide data engine that ensures accuracy and consistency in data collection and annotation, how it deployed upgraded sensors across a large fleet, how it implemented advanced dataset curation tools to mine interesting data and conquer the long tail of edge cases, and how it manages and motivates an expanding machine learning workforce. Before joining Embark, Moak previously held engineering and senior management positions at Kindred.ai and Clear Blue Technologies.