Good morning Habr!
We have nothing to add to the title of the article in our pre-notification - therefore, everyone is immediately invited under cat. We read and comment.
Developers for mobile devices will benefit from the revolutionary changes that today has to offer . It's about how this technology enhances any mobile application, namely, it provides a new level of convenience for users and allows you to actively use powerful features, for example, to provide the most accurate recommendations, or instantly discover .
This rapid development of mobile machine learning is the answer to a number of common problems that we have had time to suffer in classical machine learning. In fact, everything is obvious. Going forward, mobile applications will require faster data processing and further reduction of latency.
You may have wondered why , can't just run inference in the cloud. First, cloud technologies depend on central nodes (think of a huge data center, where both extensive data storage and large computing power are concentrated). With such a centralized approach, it is impossible to keep up with processing speeds sufficient to create smooth mobile interactions based on machine learning. The data must be processed centrally and then sent back to the devices. This approach takes time, money and does not guarantee the privacy of the data itself.
So, having outlined these main advantages of mobile machine learning, let's take a closer look at why the revolution in machine learning that is unfolding before our eyes should be of interest to you personally as a mobile developer.
Reducing latency
Mobile application developers know that increased latency can be a black mark for a program, no matter how good its features are or how respectable the brand is. Previously observed on Android devices , which often caused video and audio viewing to be out of sync. Likewise, a high-latency social network client can make communication a real torture for the user.
The implementation of machine learning on the device is becoming increasingly important precisely because of these latency issues. Imagine how social media image filters work, or location-based restaurant recommendations. In such applications, the delay must be minimal, only in this case it can work at the highest level.
As mentioned above, cloud processing can sometimes be slow, and the developer needs the latency to go to zero - only in this case, the machine learning capabilities in the mobile application will work as they should. Machine learning on devices opens up data processing capabilities that really allow you to reduce latency to almost zero.
Smartphone makers and tech giants are slowly starting to realize this. For a long time, Apple remained the flagship in this industry, developing for smartphones with the help of its Bionic system, in which the Neural Engine is implemented, which helps to drive neural networks directly on the device, while achieving .
Apple also continues to develop Core ML, its machine learning platform for mobile applications, step by step; in library added support for GPUs; Google continues to add preloaded features to its ML Kit machine learning platform. With the help of these technologies, it is possible to develop applications that allow lightning-fast processing of data, eliminating any delays and reducing the number of errors.
This combination of precision and seamless user interactions is a key metric that mobile app developers need to consider when implementing machine learning capabilities. And to guarantee such functionality, it is required .
Improved security and privacy
Another huge benefit of edge computing that cannot be overestimated is how much it improves user security and privacy. Ensuring the security and privacy of the data in the application is an integral part of the tasks of the developer, especially given the need to comply with the GDPR (General Data Protection Regulation), new European laws, which will undoubtedly affect the practice of mobile development.
Since the data does not need to be sent upstream or to the cloud for processing, there is less opportunity for cybercriminals to exploit any vulnerabilities that occur during the transfer phase; therefore, data privacy is maintained. This makes it easier for mobile app developers to comply with GDPR regulations on data security.
Machine learning on devices also enables decentralization, in much the same way as blockchain. In other words, it is more difficult for hackers to DDoS a connected network of hidden devices than to carry out the same attack on a central server. The technology could also be useful for drone operations and law enforcement.
The aforementioned smartphone chips from Apple also contribute to increased user security and privacy - for example, they can serve as the basis for Face ID. This feature of the iPhone is powered by a neural network deployed on devices that collects data on all the various representations of the user's face. Thus, the technology serves as an extremely accurate and reliable method of identification.
This and newer AI-enabled hardware will pave the way for more secure user interactions with a smartphone. In fact, developers get an extra layer of encryption to protect user data.
No internet connection required
Latency issues aside, sending data to the cloud for processing and extracting inferences requires a good internet connection. Often, especially in developed countries, there is no need to complain about the Internet. But what to do in areas where the connection is worse? When machine learning is implemented on devices, neural networks live on phones on their own. Thus, the developer can deploy the technology on any device and in any place, regardless of the quality of the connection. Plus, this approach leads to .
is one of the industries that could especially benefit from machine learning on devices, as developers can create tools that check vital signs or even provide robotic surgery without any internet connection. This technology is also useful for students who want to access lecture materials without an Internet connection, such as being in a transport tunnel.
Ultimately, machine learning on devices will provide developers with the tools to create tools that will be useful to users from all over the world, regardless of the Internet connection situation. Considering that the power of new smartphones will be at least as good as current ones, users will forget about problems with delays when they work with the application offline.
Cost reduction for your business
Machine learning on devices is also designed to save you a fortune by not having to pay external contractors to implement and support many solutions. As mentioned above, in many cases you can do without the cloud and without the Internet.
GPUs and AI-specific cloud services are the most expensive solutions you can buy. When running models on the device, you do not have to pay for all these clusters, due to the fact that today there are more and more advanced smartphones equipped with .
By avoiding the nightmarish heavy data processing that takes place between the device and the cloud, you save enormously; therefore, it is very profitable to implement machine learning solutions on devices. In addition, you save money because your application significantly reduces bandwidth requirements.
The engineers themselves also save a lot on the development process, since they do not have to assemble and maintain additional cloud infrastructure. On the contrary, it is possible to achieve more with the help of a smaller team. Thus, the planning of human resources in development teams is much more efficient.
Conclusion
Undoubtedly, in the 2010s, clouds have become a real boon, simplifying data processing. But high technology is evolving exponentially, and machine learning on devices may soon become the de facto standard not only in mobile development, but also in the Internet of Things.
With reduced latency, improved security, offline capabilities, and generally cheaper prices, it's no surprise that the biggest players in mobile development are betting big on this technology. Mobile app developers should also look into it to keep up with the times.
Source: habr.com