The emergence of Machine Learning (ML) or Artificial Intelligence (AI) has got everyone talking. A lot is being discussed and written in the newspapers about the potential of the technology in every aspect of human life and how it will change the future. Many new products are being announced and innovations are taking place in hardware, software and infrastructure levels, to best solve ML problems.
At the ground level though, ML is all about mathematical solution to a given problem via statistical and probabilistic analysis. In essence, ML converts a given problem into mathematical equation pertaining to classification, regression and such techniques. All these problems read the data, apply mathematical transformations and provide some results.
The current action in the marketplace for AI is around servers where large amount of data is consumed by the applications. These systems use powerful chipsets along with AI software framework such as Tensorflow or Python to build applications. The software framework is supplemented by several basic mathematic libraries such as Numpy to facilitate rapid algorithm development.
In the IoT (Internet of Things) world though, the problem is very different. For starters, IoT demands low power and connectivity. The computation, often times must be done on the edge device where the compute power to run Python-like framework is often not available. The algorithms must carry small footprints, so that they can be run on the device itself. A cloud platform is also required where data and events of interest could be ingested to trigger higher level functions like business processes and analytics. Needless to say, applying ML to IoT problems requires rethinking of the entire hardware and software stack that must overcome limitations above.
Today, we are announcing our new platform called QueSSence (pronounced as Q-Sense). This platform combines the best of IoT with ML and provides a one stop solution for development of intelligent IoT applications. The platform comes with ultra-low power chipset (to minimize power on the edge), AI libraries (to rapidly add machine learning to your applications) along with wide range of connectivity options (Wi-Fi, Zigbee, Bluetooth among others). The platform also comes with a secure cloud. The cloud is tightly coupled with the edge device to allow functions such as secure provisioning and over the air updates (OTA).
The platform comes with a rich set of APIs that developers can use to build applications. The ML libraries are provided in C and optimized for compute and memory requirements. The API structure of this library is very similar to high-level language such as, Python, so the developers familiar with high-level framework will find it easy to incorporate intelligence into their application. An on-board hardware accelerator provides additional compute for CPU intensive ML applications so even training could be accommodated on the edge.
To enable understanding of how one can inject intelligence into their IoT application, we have included a suite of use cases. These start with simple time series data such as gesture recognition and go all the way up to facial detection using image sensors. Ultimately successful incorporation of intelligence into any application boils down to application’s requirements in terms of performance, power and latency. We expect that there will be a set of applications where the edge alone will not be sufficient. For such applications, developers can use secure cloud and data APIs to build web applications.
Our journey is just beginning. We are proud to announce industry’s first intelligent IoT platform and feel that the applications are only limited by developer’s imagination. We want this product to be driven by the community and I encourage you to take a look at various sections on this web site to learn more. We will be hosting competitions and hackathons in coming days where developers will be able to unleash their creativity.
Original content source: Blog
Mr. Anand Joshi
Senior Director of AI products at Redpine Signals