AI Cloud: The next use of AI
By Aditya Abeysinghe
Artificial Intelligence (AI) algorithms are often memory and computation intensive due to the number of processing cycles involved. Even a model with a few source code lines will often cause several minutes to produce the output. With growing use of AI-based applications, a solution to this issue is necessary to produce the services fast with minimal lags. Therefore, use of cloud for AI processing is a new method followed by many who use AI-based functions.
Benefits of using the cloud
Different types of clouds exist. I described most of these clouds in my article on *‘Data Management Strategies in Multiple Clouds’. These clouds have different benefits and are used for different tasks.
With the use of cloud servers, scalability is possible within minimal time when memory, computation and other constraints reach limits. Scalability and elasticity can scale-up or scale-down these resources based on the situation. With auto-scaling in clouds, constraints such as computation automatically scale-up or scale-down without requiring the user to manage such situations. Therefore, costs and time for scaling, adding and removing resources is reduced with the use of cloud.
Another use of using the cloud is the ability to monitor performance of services used. Management services which monitor events enable users to monitor application performances. Alarms and other features available in these clouds notify users, when certain states change. These are useful for AI-based services because algorithms can be adjusted based on performance, under different metrics.
Reliability is another advantage of using the cloud. With the cloud, services could be hosted in multiple regions and if one hosted site fails, then the same service can be used from other regions. Clouds also provide reliability over other hosting techniques as server management is monitored by the provider and the provider serves continuously, even in case of disasters. Also, many attacks to these cloud servers are automatically reduced as the provider monitors the cloud.
Methods
The application using the AI model can be in the cloud or out of the cloud. When the application is not in the cloud, the application can be hosted on a hosting service outside the cloud or may be within an Internet of Things (IoT) device which sends data to the cloud for data processing. When the application is hosted in the cloud, all users connect to the central host to use its services. The benefit of the former method is that the application is closer to the user and is easily maintainable. For example, consider a software or app that the user is utilizing. The app or software sends data for processing to the cloud. The processed data is sent to the user and the process iterates, until the two ends send data.
Issues of using AI in the cloud
Today, most services use the cloud for storing, processing and analyzing data. Most users utilize devices which communicate with the cloud for AI processing. Therefore, during this two-way data mechanism, latency issues and attacks could occur.
Latency issues occur due to the lag of time between the requests to a service and the data sent by the service. Usually, latency occurs when the two ends are in two different regions. Latency is an issue for real-time apps which require instant processing as the sender has to wait till it gets the processed data back. Therefore, using AI models in the cloud has been raising debates in users who use IoT and other devices.
Attacks could occur when attackers intercept data sent between the cloud and the user. Attackers could drop data packets, alter and create new packets or send data to other locations. Securing cloud to user data transfer has been long researched and several methods exist. However, these methods are slow and often not suitable for processing data using AI algorithms.
*Data Management Strategies in Multiple Clouds Article: https://www.elanka.com.au/data-management-strategies-in-multiple-clouds-by-aditya-abeysinghe/
Image Courtesy: https://www.forbes.com/
