RISE Seminar 2/8/19: Data-Driven Datasets: Deep Active Learning for Autonomous Vehicles and Beyond, a talk by Adam Lesnikowski
February 8, 2019
Note: this talk has been recorded; you can watch the video on RISELab YouTube channel
Title: Data-Driven Datasets: Deep Active Learning for Autonomous Vehicles and Beyond
Speaker: Adam Lesnikowski
Affiliation: NVIDIA
Date and location: Friday, February 8, 12:30 – 1:30 pm; Wozniak Lounge (430 Soda Hall)
Abstract: Data is the source code of the software 2.0 paradigm. So why has there been a tremendous amount of focus on neural network architectures and relatively little on dataset construction in the development of modern machine learning? The speaker believes that this focus is misplaced, with the largest future gains in data-driven machine learning systems for computer vision and other applications coming from improved data set building strategies rather than architecture improvements. In particular, employing feedback from trained models allows us to iteratively build datasets and models that in many cases leads to substantial improvements in performance with less labelled examples. The speaker will present several cases studies related to autonomous vehicle at NVIDIA where this paradigm has been exploited, and speculate on current and future areas of challenges and opportunities.
Bio: Adam is a machine learning scientist and former senior software perception engineer at NVIDIA, focusing on applied research problems related to autonomous vehicle data collection. Previous to NVIDIA, he founded and was a CEO of a startup in the computer vision and navigation space. Prior to that, Adam worked on problems in mathematical logic at Berkeley for grad school and was at Harvard undergrad for math. In his spare time, he enjoys travel, all things science, technology and math, games, running, science fiction, and dogs.