This video is part of the TechXchange: TinyML: Machine Learning for Small Platforms.
The term tinyML represents an idea rather than a framework like TensorFlow. It normally refers to artificial intelligence/machine learning (AI/ML) on the edge, where microprocessors have more limited capabilities and low power limits.
The tinyML Foundation fosters innovation in this area, and it recently held its annual summit. If you weren't able to attend in person, you're in luck, because like many trade shows these days, the presentations are being recorded and posted online.
I started this article off with the 2023 keynote by Ian Bratt, Fellow and Senior Director, Central Technology Group, Arm, entitled "tinyML: From Concept to Reality" (see video above). He notes that "tinyML is at a tipping point. This community has come together to form a stable technology foundation, enabled by standardization of software and methodologies, which will enable tinyML to scale at a level that’s never been seen before."
I've collected links to most of the presentations and embedded a couple of my favorites below. TinyML is a technology that we're covering more and more at Electronic Design because of our embedded focus as well as the fact that so many applications are dealing with low-power and mobile solutions.
- Machine Learning Sensor Certifications and tinyML edu Update
- A perspective on the trajectory from custom intelligent sensors to broad market adoption of smart platforms
- Deploying Visual AI Solutions in the Retail Industry
- Arm Ethos-U support in TVM ML framework
- Using tinyML and Sound Event Detection for weld anomaly detection in Manufacturing
- End-to-End MLOp system for pre-clinical medical research
- Low-Energy Physiologic Biomarker Machine-Learning Inference on a Wearable with a GAP9 RISCV Processor
- Personal Computing devices use-case and applications enabled by Smart Sensors
- How a consumer goods company leverages Qeexo’s AutoML to accelerate data. Science adoption and value
- Multi-Lingual Digital Assistance on Edge Devices
- Enhancing neural processing units with digital in-memory computing
- Responsible Design of Edge AI: A Pattern Approach for Detecting and Mitigating Bias
- Designing Multi-Model Smart Human Machine Interfaces with Microcontrollers
- How can we find real uses for tinyML?
- Low Power Radar Sensors and TinyML for Embedded Gesture Recognition and Non-Contact Vital Sign Monitoring
And here are a few of my favorites.
Why TinyML Applications Fail: An examination of common challenges and issues encountered for real-world projects
TinyML applications can be a challenge, as is the case with any embedded application that has limited processing power, storage, and overall power. Then again, it's sometimes a matter of finding out what fits within these limitations.
Tiny spiking AI for the sensor-edge
Spiking neural networks offer many advantages, including compact size, lower computation requirements than deep neural networks, plus being easier to train in the wild. There are challenges, though, which is why spiking neural networks coexist rather than replace other machine-learning approaches.
tinyML application throwdown: What application area has the most potential?
Want to know what applications can take advantage of tinyML? Check out the video to learn more.
Check out more videos and articles in the TechXchange: TinyML: Machine Learning for Small Platforms.
