ARL Science Talk: Dr. Ethan Stump - Parsimonious Online Learning with Kernels

September 05, 2017

Because the world is so ever changing and unpredictable, robots must learn at the same time as they act.

In this ARL Science Talk, Dr. Ethan Stump explains how a mobile robot can learn about its environment - and how to move in that environment - at the same time as doing the movements. Granted, the last few years have seen an explosion of rapid developments in applying machine learning, particularly neural networks, to practical problems. However, many future problems - including military problems - will not be solved by current approaches due to the dynamic nature of the modern battlefield. Robotics scientist Dr. Ethan Stump discusses general aspects of the machine learning problem that will be critical for enabling autonomous systems (robots), and presents an approach he and his colleagues have been developing to address these aspects using online learning with reproducing kernels. Though this technique reaches back to an earlier wave of development in machine learning, it provides a important complement to current practices centered on neural networks because it enables data-efficient learning and admits intuitive models.


Last Update / Reviewed: September 5, 2017