Below is a guest post by the CEO of SensiML, a QuickLogic partner specializing in embedded machine learning. As for fulfillment, we finally got the paperwork sorted and will start shipping by September 23rd. You will receive a shipping confirmation email with a tracking number when your order ships. Thank you for your patience!
When QuickLogic launched the QuickFeather campaign earlier this summer, it referenced the coming integration of the SensiML AI toolkit with this uniquely flexible and open source QuickFeather dev kit. We are pleased to report the integration is complete and we are making SensiML Analytics Toolkit available as a specially priced bundled option for Crowd Supply QuickFeather backers.
Anyone who has ever created sensor data processing applications for
embedded microcontrollers knows this activity can be extremely
challenging and time-consuming. True, while simple threshold rules
(like a temperature alarm or linear motion travel limit) can be
handled with a single
IF...THEN, the more interesting problems in
embedded sensor processing require much more; and typically demand far
greater attention to algorithm efficiency than cloud-based machine
learning since computing resources, memory, and power are often quite
The alternative is to forego building intelligence intro the embedded processor and use the IoT edge device as simply a data acquisition node to capture and forward raw sensor data for processing elsewhere (cloud, PC, smartphone, etc.). This "dumb" IoT node workaround is acceptable in some use cases, but more often is the case where the need for real-time responsiveness, low network utilization, and/or data privacy demand local processing at the extreme edge. In most cases, autonomous smart devices having edge inferencing built-in will provide a better application performance and user experience overall.
Just a few examples of applications for this type of sensor processing include:
If your application calls for any of the above or similar pattern-recognition type classification and the performance of edge sensor processing, then SensiML may be just the solution for your project.
Whereas traditional methods for constructing embedded sensor processing algorithms rely largely on time-intensive data science, signal processing expertise, and hand-coded optimization, SensiML automates the creation of sensor processing algorithms by training machine learning algorithms by example using labeled training data you provide within the tool. In this way, your efforts can be focused on the application logic and data interpretation outcomes rather than the nuances of AI and efficient embedded code implementation.
Using the Data Capture Lab component of SensiML Analytics Toolkit, you collect example sensor data and then label this data with the help of toolkit video synchronization, predictive auto-labeling, and metadata annotation that can be used to further segment data during model creation.
Following this step, you can rapidly create library or binary code for the QuickFeather ready for flashing to the device for test once you are satisfied with the predictive emulation results within the development tool. From start to finish, creation of working predictive algorithms that run locally on the QuickFeather can be built in a matter of hours once you have created and labeled a training dataset.
We are offering a bundled option to backers of the QuickFeather open source IoT dev kit, that includes both the QuickFeather dev kit and a license to SensiML Starter Edition allowing you to build smart IoT sensor algorithms quickly and easily with this kit.
SensiML Starter Edition includes:
For more information on SensiML Analytics Toolkit, visit our website at https://sensiml.com. If you’ve already placed your QuickFeather order and would like to upgrade to the QuickFeather + SensiML AI Bundle, contact Crowd Supply support.
Collecting 6-axis accel/gyro sensor data for boxing punch recognition application
Showing graphical visualization of algorithm performance results for automatically created AI model
A growing repository of example datasets across a range of IoT sensor applications