5 minute read
Tiny ML
tinyML is the next wave of Machine Learning that is enabling low power embedded systems to do what was once only possible in the domain of high performance cloud computing. Its fuelling the emerging field of Edge Machine Learning which makes it possible to run deep learning models directly at the edge on 32bit micro controllers. This has been made possible by breakthrough research into the optimization and deployment of convolutional neural networks pioneered by the TensorFlow Lite Team at Google.
The focus of tinyML is to not just deploying deep learning neural network models on micro controllers but to also keeping the power consumption as low as possible. Newer systems and embedded AI accelerators are coming to the market at an unprecedented pace, bringing support for ever increasing model complexity at reducing levels of power consumption.
This has brought the worlds of embedded systems and Artificial Intelligence together in turn activating new use cases and opportunities that were previously not feasible.
Internet of Things or IoT evolved from adding connectivity to embedded systems to facilitate the capturing and forwarding of raw data about processes to the cloud, where machine learning would usually be used to derive insights. This consumes power and bandwidth and introduces a latency in the processing of the data into results.
By leveraging unused computing capacity in these IoT devices to deploy tinyML models to derive insights directly from the data on the device itself, IoT becomes the Artificial Intelligence of Things or AIoT. The instant benefits are lower bandwidth and privacy as the raw data doesn't need to leave the device and its possible to send just the insights to the cloud via low cost, low bandwidth and of course low power data connections. The added benefits are longer periods between replacing batteries and immediate intelligent responses to the environment.
Popular applications include detecting and classifying objects in front of tiny low powered cameras, keyword spotting and figuring what sounds mean in real time from microphones. It doesn’t stop there and its possible to work with any kind of sensor to find interesting applications such building a low cost highly accurate artificial nose that runs on batteries as an interesting example of the kinds of new use cases made possible by tinyML.
The growing use cases span all industries and allow for businesses to implement predictive maintenance and demand optimisation in real time directly on the ground. In a quick response driven world, the agility provided by tinyML not only helps the bottom line but helps identify and reduce and prevent process bottlenecks proactively in a sustainable way that also protects energy resources.
It doesn’t stop with the technology itself, tinyML is driven and supported by the TinyML Foundation and its partner hardware and software vendors who are all working together with a common purpose of changing the world of AI and bringing true intelligence to every kind of device around you while also empowering you to be able to do it yourself. For the first time in history big tech and industry are working together with the community and end users from school learners and hobbyists to professionals alike to create and nurture new opportunities.
TinyML embodies this approach as more than just a technology but a philosophy and a movement with the goal of also uplifting the world through initiatives like TinyML for Good to drive sustainability and empower communities and democratize this disrupting technology.
This is achieved through Meetup Groups, TinyML Talks, open access to information and resources all driven by the TinyML Foundations Strategic Partners forming a vibrant, diverse and thriving global community who all support each other with the shared goal.
The pandemic has brought the world closer together and never before has it been more apparent that everyone is a global citizen with a part to play. The developing world is no exception as technology will only help solve problems if its affordable and accessible. TinyML supports this through TinyML4D run out of Harvard that is making the technology open and accessible to all through the creation and dissemination of courseware tailored to support the developing world.
In Africa the ecosystem is growing steadily having started in Northern Africa through the efforts of early inspirational pioneers who saw the need to develop skills as way to provide economic sustainability for their communities.
We have only just begun in Africa and there is a still a lot more work to do to spread it to all parts of Africa. In South Africa there are many challenges that need to be overcome which can be solved through job creation and entrepreneurship which in turn will be possible through skills development.
This is driven by initiatives such ARM E 3 NGAGE and Coretex Hub where skills are being nurtured at grassroots levels in STEM and Robotics and the ARM E 3 NGAGE Embedded Learning Challenge to equip the next generation of embedded engineers and incubate entrepreneurship.
Most importantly TinyML is open to all regardless of background whether you are a developer, academic, maker, AI expert or just someone who is curious and wants to find a way into AI as a career path don’t know where to start all are welcome all to join the movement and the Meetup Group in your area or the ARM E 3 NGAGE Embedded Learning Challenge.
We hope to see you there.
To learn more visit: http://www.tinyml.org/ or https://www. thecortexhub.africa/armlearning-challenge