Håkan Ohlgren, CTO and Chief architect shares some insights in the power of AI.
These days everyone is talking about AI, AI on the Edge and Deep Learning. Media is painting a scary picture of the future to come and probably – at least a part of you – agree that the unknown feels a bit scary – right? Personally, I can sometimes get afraid as I get an insight in the power of AI. But then my curiosity takes over and I start seeing the possibilities all around me.
The thing is that AI is not just about a super computer network gathering and processing tremendous amount of data with the ability to “control” the population – as media sometimes declares. It is also about optimizing the power consumption in a sensor network, based on user behavior. It is also about reducing component cost in a device. It is also about reducing the overall power consumption in a city, as the system learns the behavior of the population. Yes – AI can be used in autonomous driving vehicles, and it can also help gathering data from a mine or underground/water cave, using a self-flying drone. What if data from your sensor system that is intended to monitor a specific machine, can also be used to predict an earthquake?
What is deep learning use for?
Deep learning can be used to process the data of a huge number of sources in order to learn typical patterns, identify normal behavior etc. It can then identify abnormal behavior and scenarios and warn for a possibly hazardous situation or just prevent a machine from getting damaged or worse – a person from getting hurt. The more data available, the better the predictions.
A concrete example of Artificial Intelligence in a small device
But AI is not only about Deep Learning. Let’s look at a small device with limited battery capacity. It connects to the cloud through cellular connectivity. The connectivity consumes quite much energy and each time data is exchanged, the battery longevity is significantly reduced. In this example, the data exchanged is used to authorize the user and to give the cloud sensor data in order for it to track the device and act upon its position or unintended usage. Now what if the device itself could manage the authorization and understand by whom and how the device is used. By implementing an Edge AI engine in the device, it can learn the normal behavior of one or several users. Rather than using various expensive Bio-metric sensors, the AI engine can learn about the user by monitoring time of day, motion and tremor patterns using inexpensive MEMS sensors. Based on the “normal behavior” model, any significant deviations can trigger a connection to the cloud and/or deactivation of the device.
This short example is a typical application where Edge AI can be used to reduce cost and to improve the battery longevity of a device. Of course, a complex AI engine consumes memory and processing power and thus the cost might increase. But depending on the application, the algorithm might not have to be that complicated.
At WSI we meet a wide variety of customers with an even wider range of products. And we see a clear trend of connecting more devices to the cloud (IoT). We also see an increased interest in AI where customers are curious about what AI could do for them. At WSI we can help you to systemize, develop and manufacture your product. And with our extensive partner network, we can find a suitable AI partner that would boost your application if we find that AI is applicable for you. As an official design partner for NVIDIA, we have access to the leading companies working with Deep Learning using the NVIDIA CUDA GPU processing platform, which is used in products ranging from self-flying drones, Autonomous vehicles to the server implementations used by the big Cloud vendors offering AI services. But we also have partners working with Edge AI algorithms in small and low cost embedded systems.
So let’s ask yourself again – Artificial Intelligence is not for me – right? Well I think that it might well be!