AI algorithms are processed locally, either directly on the device or on the server near the device. The algorithms utilize the data generated by the devices themselves. Devices can make independent decisions in a matter of milliseconds without having to connect to the Internet nor the cloud. Edge AI has almost no limits when it comes to potential use cases. Edge AI solutions and applications vary from smartwatches to production lines and from logistics to smart buildings and cities.
How does Edge AI work, what type of business benefits does it bring and how can you get started with Edge AI? Read this page and find out - let's begin our journey to the edge!
👉 Edge Computing
Edge computing consists of multiple techniques that bring data collection, analysis, and processing to the edge of the network. This means that the computing power and data storage are located where the actual data collection happens. What is meant by the network edge? Well, depends on the use case - it could be a mobile phone, IoT-device, self-driving car or even a cell tower. Read more about edge computing from Samu’s blog.
👉 Artificial Intelligence
Broadly speaking, in Artificial Intelligence a machine mimics human reasoning: such as understanding languages and problem solving. Artificial intelligence can be seen as advanced analytics, (often based on machine learning) combined with automation. This pragmatic definition covers all current AI applications.
You can think of Edge AI as analytics that takes place locally and utilizes advanced analytics methods (such as machine learning and artificial intelligence), edge computing techniques (such as machine vision, video analytics, and sensor fusion) and requires suitable hardware and electronics (which enable edge computing). In addition, location intelligence methods are often required to make Edge AI happen.
Edge AI devices include smart speakers, smart phones, laptops, robots, self-driven cars, drones, and surveillance cameras that use video analytics.
Although most of the analysis and decisions made by machines already happen on the edge, the greatest benefits are obtained when the findings produced by the devices are linked to business processes. Therefore, modern data platforms, capable of handling large amounts of location and streaming data, are also needed to enable real-time computing.
Tip for the busy ones: there is a handy navigation pane on the left side. Use it to move straight to the section that interests you. 👀
In a typical machine learning setting, we start by training a model for a specific task on a suitable dataset. Training the model basically means that it is programmed to find patterns in the training dataset and then evaluated on a test dataset to validate its performance on other unseen datasets, which should have similar properties to the ones that the model is trained on.
Once the model is trained, it is deployed or in other words "put to production".
It can now be used for inference in a specific context, for example as a microservice. Inference refers to the process of using a trained machine learning algorithm to make predictions.
Once the model works as wanted, predictions produced by the model can be utilized in improving business processes. Typically, the model works via an API. The model output is then either communicated to another software component, or in some cases, visualized on the application front-end for the end user.
Cloud is not enough
The rise of cheap computing and data storage resources with cloud infrastructure has given new opportunities to leverage machine learning at scale. However, this comes at the cost of latency and data transfer challenges due to bandwidth limitations.
Training a machine learning model is a computationally expensive task well suited for cloud-based environment, whereas inference requires relatively low computing resources.
Given the new trends of Industry 4.0, autonomous systems and intelligent IoT devices, the old paradigm of executing inference in cloud is starting to get increasingly less suitable as needs for real-time predictions grow.
If a machine learning model lives in the cloud, we first need to transfer the required data (inputs) from the end-device, which it then uses to predict the outputs. This requires a reliable connection and if we assume that the amount of data is large, the transfer can be slow or in some cases impossible. If the data transfer fails, the model is useless.
In the case of successful data transfer, we still need to deal with latency. The model naturally has some inference time, but the predictions also need to be communicated back to the end-device. It's not hard to imagine, that in mission-critical applications, where low latency is essential, this type of approach fails.
Computationally more powerful edge devices have enabled a new way of performing machine learning and artificial intelligence - Edge AI.
In the traditional setting the inference is executed in a cloud computing platform.
With Edge AI, the model works in the edge device without requiring connection to the outside world at all times. The process of training a model on a consolidated dataset and then deploying it to production is still similar to cloud computing though. This approach can be problematic for multiple reasons.
First, it requires building a dataset by transferring the data from the devices to a cloud database. This is problematic due to bandwidth limitations. Second, data from one device can not be used to predict outcomes from other devices reliably.
Finally, collecting and storing a centralized dataset is tricky from a privacy perspective. Legislative limitations such as GDPR are creating significant barriers to training machine learning models. Moreover, the centralized database is a lucrative target for attackers. Therefore, the popular statement that edge computing alone answers to privacy concerns is false.
For tackling the above problems, federated learning is a viable solution.
Federated learning is a method for training a machine learning model on multiple client devices without having access to the data itself.
The models are trained locally on the devices and only the model updates are sent back to the central server, which then aggregates the updates and sends the updated model back to the client devices. This allows for hyper-personalization - while preserving privacy.
Edge computing is not going to completely replace cloud computing, rather it's going to work in conjunction with it.
There are still multiple applications, where cloud-based machine learning performs better, and with basic Edge AI the models still need to be trained in cloud-based environments. In general, if the applications can tolerate cloud-based latencies or if the inference can be executed directly in the cloud, cloud computing is a better option.
One of the most promising Edge AI use cases is manufacturing quality control. Advanced machine vision (video analytics), an example of Industrial Edge AI, can monitor product quality tirelessly, reliably and with great precision.
Video analytics can detect even the smallest quality deviations that are almost impossible to notice with the human eye.
Production automation requires advanced analytics, for example in the prediction of equipment failures. Analyzing the data from the sensors and detecting abnormalities in near real-time makes it possible to shut the device off before it breaks. This can save you from significant hardware damages or even injuries. Automatic analysis of material flows by video analysis, for example, is also a promising use case.
Passenger air crafts have been highly automated for a long time. Real-time analysis of data collected from sensors can further improve flight safety.
While fully autonomous and fully unmanned ships may not become a reality until years from now, modern ships already have a lot of advanced data analytics.
Edge AI technology can also be used, for example, to calculate passenger numbers and to locate fast vehicles with extreme accuracy. In train traffic, more accurate positioning is the first step and a prerequisite towards autonomous rail traffic.
A smart grid produces a huge amount of data. A truly smart grid enables demand elasticity, consumption monitoring and forecasting, renewable energy utilization and decentralized energy production. However, a smart grid requires communication between devices, and therefore transferring data through a traditional cloud service might not be the best alternative.
Large retail chains have been doing customer analytics for a long time. The analytics is currently largely based on an analysis of completed purchases, i.e. receipt data. Although good results can be achieved with this method, the receipt data does not tell you everything. It doesn’t tell you how people move around the store, how happy they are, what they stop to watch, etc. Video analytics analyses fully anonymized data extracted from a video image and provides an understanding of people’s purchasing behaviour that can improve customer service and the overall shopping experience.
If you read this far, you probably already wonder how Edge AI solutions could work as a part of your business.
Success in Edge AI depends on having courage and an open-minded attitude: be agile, dare to fail, learn from your mistakes, and scale your success into new business processes and service development.
Book a free half-hour telephone conversation with us, so we can brainstorm together how Edge AI solutions could benefit your business, what kind of use cases are feasible in your situation and what kind of challenges different implementations might have. 👇
