TEARDOWN 2019: Measuring embedded machine learning
Talk by Alasdair Allan
While machine learning is traditionally associated with heavy-duty, power-hungry processors, the future of machine learning is on the edge and on small, embedded devices that can run for a year or more on a single coin cell battery. Deep learning can be very energy-efficient, and allows us to makes sense of sensor data. It can turn raw accelerometer data into information about whether a machine is working well or malfunctioning, or recognize voice commands from a microphone feed. The ability to run trained networks “at the edge” nearer the data without the cloud - or even without even a network connection - means that we can interpret sensor data in real-time, pulling signal from the data without storing potentially privacy infringing data in the cloud. This talk shows you how to use machine learning on your own problems, whether you’re using your laptop, a Raspberry Pi, or an ARM Cortex M micro-controller. It looks at the latest hardware intended to speed up machine learning inferencing on the edge, and gives benchmarks as to which platform is fastest.
Bio: Alasdair is a scientist, author, hacker, maker, and journalist. An expert on the Internet of Things and sensor systems, he’s famous for hacking hotel radios, deploying mesh networked sensors through the Moscone Center during Google I/O, and for being behind one of the first big mobile privacy scandals when, back in 2011, he revealed that Apple’s iPhone was tracking user location constantly. He has written eight books, and writes regularly for Hackster.io, Hackaday, and other outlets. A former astronomer, he also built a peer-to-peer autonomous telescope network that detected what was, at the time, the most distant object ever discovered.