Today SensiML took a leadership position in AI/ML tools for the IoT edge by announcing our new Open Source Initiative. Our goal with this initiative is to accelerate the adoption of TinyML technology smart sensing IoT applications for real-world products by committing to deliver open and transparent ingredients as required to support commercial IoT products involving AI technology. This is both a formal initiative bringing together existing open elements of the SensiML Analytics Toolkit as well as a plan for extending this to incorporate additional open components that offer design flexibility and transparency during the development process.
The potential for intelligent IoT endpoints is practically unlimited, and we are big believers that the TinyML approach is the wave of the future. TinyML leverages the underutilized processing power of the embedded endpoints to provide distributed analytics where sensor data processing can be partitioned optimally for improved power, performance, responsiveness, and data security/privacy. SensiML is uniquely positioned to support this approach as our tools enable edge IoT devices to use AI/ML capabilities to make smart, local decisions with efficient AutoML models that are optimized to fit the power and compute constraints of the smallest embedded platforms.
That said, the implementation of AI at the extreme IoT edge is quite new and already developers struggle with the “black-box” aspects of AI whether in the cloud, network edge, or IoT edge. Add to this the take-it-or-leave-it aspect of many AutoML tools that lack both transparency AND flexibility in generated model code, and it’s not surprising that developers are slow to adopt for commercial products they must support in the market. The ability to create supportable, commercial quality designs quickly and efficiently with or without data science expertise is the difference SensiML brings to TinyML. With our open source initiative, we just supercharged that strength by providing new open source utilities, protocols, datasets, and reference code to add customizability, efficiency, and visibility from top to bottom.
* SensiML Open Source Embedded SDK to be released Summer 2021
SensiML Open Gateway: A fully open source, user-extensible, multi-protocol application for connecting embedded IoT sensor devices to SensiML’s data collection tools for train and test data collection (SensiML Data Capture Lab and SensiML TestApp), as well as connectivity to cloud IoT platforms and other endpoints.
SensiML Open Data Interfaces: With support for both simple streaming binary output and full IoT device command/control using MQTT-SN, SensiML’s application-level protocols for bringing sensor data into our tools are published and available using open-source reference code examples.
SensiML Open Mobile Test App: The company’s Android variant of the device testing tool SensiML TestApp is offered as open source software to support flexible options for data testing in mobile settings.
SensiML Open Source Embedded SDK (coming later this summer) – The full library of SensiML segmenters, transforms, features, and classifiers as implemented by the AutoML and Python-based SensiML Analytics Toolkit Notebook will become available in open source format. The SensiML Open Source Embedded SDK will provide full insight and transparency of model operation and leverage the collective expertise of our partners, users, and the embedded and AI community at large to make available updates and platform specific optimizations over time.
It is our belief that these and ongoing open elements will aid commercial product developers to innovate their IoT sensing applications with:
- Reduced development time and rapid prototyping of innovative smart IoT product concepts using data-driven supervised and unsupervised ML model development
- An ability to add AI/ML functionality to products without in-depth data science experience or expertise utilizing SensiML’s AutoML modeling engine
- Faster time-to-market with differentiated intelligent IoT devices that remove dependence on cloud and smartphone AI processing to deliver sensor insights
We hope you’ll agree and give our AI tool suite for the edge a look for your next innovative IoT sensing product if you aren’t already!
Real-world IoT products must be supportable and supportability demands firmware that developers can understand and adapt as needed. AI models are no exception here.