The ability to do machine learning problems on the edge spans a wide spectrum of hardware platforms; promulgating advanced machine learning and artificial intelligence work on smaller ecosystems ranging from Nvidia Jetsons to Raspberry Pi single board computers to tiny, frugal devices that have at-memory computer and native processing neural network architectures contained on quarter-sized boards. When combining the capabilities of modern frugal devices and the ability to compress models, there is an opportunity to innovate and create within a frugal ecosystem.

This presentation will showcase merging the frugality of ML-enabled boards and the ability to leverage compressed computer vision models to detect potential wear and tear in ship hardware/components. The ability to build and quantize machine learning models to deploy to a microcontroller can offer a light payload; therefore, enabling analysts and users to conduct machine learning inferencing in limited access areas. Moreover, these low-visibility and disposable hardware can become the preferred and ideal sensing mechanism in austere and unique operational environments.

The audience will walk away with a clear understanding how necessity, coupled with frugality, can become a powerful way to innovate. They will see how machine learning models that have been quantized and compressed to work on frugal, lightweight devices as payloads for numerous applications.