If you’re developing AI-based consumer wearable or industrial IoT products, you’ll likely be evaluating various software tools to help implement AI on your wearable or IoT device. The options range from cloud-based analytics which offers nearly limitless computing resources but introduces communication issues such as latency, bandwidth constraints, fault tolerance, and security – to local embedded development tools which rely upon data scientists and firmware engineers to hand-code algorithms at significant cost, dev time, and schedule impact.
The ideal solution would automate the algorithm development task with efficient code that can execute in real-time on the embedded device offering low cost, low power consumption, low latency, high fault tolerance, and security. Fortunately, there are now tools on the market which support just such a solution by bringing powerful AI capabilities to local endpoint applications implemented using MCUs. Those same tools not only create optimized solutions but also support easy implementation along with rapid prototyping and model evaluation, eliminating the need for huge teams of data scientists and engineers and/or allowing such experts huge leaps in productivity by automating many of the steps in the workflow.
For example, the SensiML Analytics Toolkit provides an end-to-end AI development platform spanning data collection, labeling, algorithm and firmware auto-generation, and testing. Using this type of platform enables developers to quickly and easily generate application-specific pattern recognition code that transforms sensors into smart, actionable event detectors. Some of the specific benefits of using this type of tool include:
– Creating optimized AI code without data science expertise
– Generating code that can run on MCUs rather than CPUs
– Making designs flexible and easily extensible
– Supporting a full end-to-end workflow
– Delivering “production-grade” capabilities such as multi-user, multi-project, and dataset management
– Getting significant time-to-market gains (up to 5x faster than AI expert hand coding)
The toolkit has been designed to support multiple hardware evaluation platforms including the QuickLogic Merced and Chilkat platforms, Raspberry Pi, the ST Sensor Tile and the Nordic Thingy, making evaluation just that much quicker and easier.
To further lower the barriers to entry, at SensiML we recently decided to offer a trial version of our Analytics Toolkit for free. The trial version includes:
– Data Collection and Labeling (using sample datasets)
– Automated ML Algorithm Creation
– Algorithm Performance Visualization
– Auto-Generation of Optimized Device Code
– Embedded Binary Executable Output (limited number of classification results per device power cycle)
– Device Test/Validation Application
With this version, you can easily explore the design flow used to quickly create AI-based IoT applications and gain a deeper understanding of how the SensiML Analytics Toolkit can help your own application development. We’re confident that once you try the tools, you’ll want to start using them on a regular basis.