Talk

ML on the Edge: Tradeoffs and Requirements

Watch Now

28:55

EOC 2023

Login to Watch

Kate Stewart

Kate Stewart works with the safety, security and license compliance communities to advance the adoption of best practices into embedded open source projects. Kate was one of the founders of SPDX, and is currently the specification coordinator. Since joining The Linux Foundation, she has launched the ELISA and Zephyr Projects, among others, and oversees other embedded projects hosted by the Linux Foundation. With over 30 years of experience in the software industry, she has held a variety of roles and worked as a developer in Canada, Australia, and the US and for the last 20 years has managed software development teams in the US, Canada, UK, India, and China. She received her Master's in computer science from University of Waterloo and Bachelor's of computer science from the University of Manitoba.

View full profile

Topics Covered

AI/ML RTOS Security Open Source

Talk

ML on the Edge: Tradeoffs and Requirements

Kate Stewart

28:55 EOC 2023

Kate Stewart

28:55

Over the last few years, we're starting to see machine learning be more effectively deployed closer to where data is collected in embedded systems. These end point devices may be resource constrained though, either in terms of power, memory or communication capabilities - sometimes all three. Being able to apply machine learning on these end point devices is possible, and enables system-wide efficiencies to be realized. This talk will explore the requirements and tradeoffs for such systems to be considered when using the Zephyr RTOS and Tensorflow Lite for Embedded Microcontrollers projects.

1 / 4

Please log in or create an account to test your knowledge and see the answers.

According to Kate Stewart, what is the primary benefit of running machine learning on resource-constrained edge devices rather than sending all data to the cloud?

A It eliminates the need for any cloud services for the product lifecycle.

B It enables system-wide efficiencies by processing data where it is collected, reducing communication overhead and only sending important events upstream.

C It allows devices to run arbitrarily large, complex models locally without regard to power or memory.

D It guarantees that devices will never need security updates because nothing is transmitted.

E It primarily reduces development time since models are easier to train on-device.

What approach does Kate recommend for keeping the memory footprint small when building Zephyr images for ML-enabled edge devices?

A Include a broad set of drivers and enable features at runtime so you can selectively use them later.

B Use DeviceTree together with Kconfig to describe hardware and configure only the components you actually need, producing a lightweight static footprint.

C Port a full Linux kernel and strip services after the image is built to reduce size.

D Rely on dynamic linking of libraries on the device to reduce image size.

E Use a high-level virtual machine on the device so components can be loaded as packages at runtime.

How does Zephyr’s SBOM and build-tracing capability help teams managing long-lived edge devices with potential vulnerabilities?

A The SBOM automatically patches vulnerable components in deployed devices without developer action.

B Build SBOMs link source files to the final image so you can prove whether vulnerable files were actually included, avoiding unnecessary field updates when not affected.

C SBOMs allow vendors to hide proprietary modules so attackers cannot find vulnerabilities.

D Once an SBOM exists, vulnerabilities no longer need to be monitored because all issues are resolved at build time.

E SBOMs replace the need for long-term stable releases and backported security fixes.

Which of the following best captures Zephyr’s approach to connectivity and hardware support for ML on the edge, as described by Kate Stewart?

A Zephyr focuses only on Bluetooth Low Energy and ignores other protocols to keep the OS minimal.

B Zephyr provides a wide variety of wired and wireless connectivity options plus support for many cores, sensors, and boards so devices can handle different communication needs and supply‑chain variability.

C Zephyr requires a single fixed hardware reference design to ensure consistent ML behavior across deployments.

D Zephyr removes networking entirely for edge ML, since on-device inference eliminates the need to transmit data.

E Zephyr only supports ARM Cortex‑M devices because other architectures are too large for edge ML.

Formatting help

italics	surround text with asterisks
bold	surround text with two asterisks
hyperlink	[hyperlink](https://example.com) or just a bare URL
inline code	surround text with single `backticks`
code block (multi-line)	wrap on its own lines with three backticks: ``` your code here ```
~~strikethrough~~	surround text with ~~two tilde characters~~
quote	prefix with >

Upvotes Newest Oldest

No comments or questions yet. Be the first to start the conversation!

Partners & Sponsors

STMicroelectronics

Amazon AWS

Golioth

The Qt Company

Memfault

NXP

SEGGER Microcontroller

PE Micro

PX5

Become a Sponsor