that are required for machine learning are available. Machine learning is also frequently used because problems such as Predictive maintenanceare so complex that they are very difficult or even impossible to manage using conventional programming methods1.
Today, there are technologies that make it possible to operate machine learning even without in-depth expert knowledge. So-called Automated Machine Learning (AutoML) Systems automate expert tasks and make machine learning accessible to developers.
Cloud-based solutions for rapid use of machine learning applications
In addition to data preparation, one of the key challenges of a machine learning project is the Deployment. For example, input data is generated more frequently and in greater quantities under production conditions. It is therefore important to have a Scalable and stable system to be made available. Many cloud providers already offer standardized machine learning solutions2that offer scalability, stability and AutoML. However, it is often unclear which provider is right for the problem. This article shows how to choose the suitable provider for your own use case finds.
Requirements for a machine learning solution in the cloud
A combination of expert interviews and research into current studies has shown that a solution in the machine learning and cloud environment essentially has the following properties should have:
- The machine learning process, or model creation, is automated.
- The trained model was to be delivered in a public cloud.
It is sufficient to fulfill these properties, AutoML for the creation of the model and then to use it in a Cloude.g. to a public cloud. Only the training data still has to be collected manually.
The following illustration shows a process that fulfills these characteristics:
The advantage of using a pipeline with the above structure is that the flexibility of a cloud solution can be fully utilized even with little expert knowledge in the field of machine learning. Through the use of AutoML can thus faster results can be achieved. The deeper the expert knowledge in the field of machine learning, the more accurate the results can be and, if necessary, the faster they can be achieved.
In a study we conducted Market analysis the following solutions emerged, which already implement the above pipeline:
Decision-making - which cloud solution suits the use case?
The question now arises as to which provider is best suited to your own use case. To answer this question, we have created a Rating system has been developed. This evaluates the building blocks of the above pipeline and consists of three parts:
- General (General): Criteria relating to the application itself are taken into account here, such as the number of components in this system.
- Machine LearningCriteria that evaluate the machine learning part of the solution are taken into account here.
- Cloud / Deployment: The deployment part of the solution is evaluated in the Cloud section.
The following illustration shows this structure.
These criteria were given weightings that reflect expert knowledge and findings from current studies. We have Can be used in an Excel sheet made.
This enables everyone to find the right cloud provider for their individual cloud-based AutoML solution.
Happy scoring!
