Skip to main contentSkip to page footer

 |  Blog

Automating machine learning projects with AutoML

With the recent successes in the development of generative AI such as ChatGPT, the term "artificial intelligence" has finally broken through to the general public and is becoming increasingly popular in society. In conjunction with the advancing digitalization and the growing amount of data, it is no surprise that more and more companies are considering or discussing the use of AI.

However, the implementation of an AI project is often not easy due to the high adaptability and complexity of the problems to be solved and requires time and expertise that only a limited number of experts have. It is therefore very important to make the best possible use of existing expertise in order to achieve a satisfactory result or proof of concept quickly.

The problem of limited expertise in the implementation of AI projects could now be solved by "Automated Machine Learning" (hereinafter AutoML). AutoML is an emerging field in the field of artificial intelligence that deals with the automation of the machine learning workflow. The aim of AutoML is to automate as many steps as possible in a machine learning project in order to shorten the development time and make it more efficient. To this end, automation tools and algorithms are used to automate the manual selection and configuration of algorithms and hyperparameters.

 

In principle, a machine learning workflow supported by AutoML comprises the same steps as a classic machine learning workflow. In addition, individual steps are automated. However, the automation of individual steps remains a challenge and cannot always be achieved to the same extent. For optimization, the steps of model selection and hyperparameter optimization are the most suitable, as they are non-specific and independent of the use case and are therefore considered the core of AutoML. In addition, there are efforts to automate further steps such as data cleansing, feature creation and selection, and explainability of predictions. AutoML can also support data scientists in some of these steps, especially when little or no domain-specific knowledge is required. For example, standard mechanical tasks such as the standardization of column entries can be performed during data collection and preparation.

 

AutoML solutions can be divided into three categories:

  1. Standalone code packages such as AutoGluon, H20 AutoML and TPOT require programming skills in Python.
  2. AutoML solutions in cloud services such as Azure Machine Learning, Amazon SageMaker or Google's Vertex AI are easy to use and offer both no-/low-code user interfaces and the ability to process your own notebooks with cloud resources.
  3. Specialized data science platforms such as Dataiku, H20, RapidMiner or DataRobot offer a user-friendly graphical interface and can be tailored to specific use cases. As a rule, the functions and support in the overall context of MLOps are more extensive for data science platforms.

 

Thanks to its many years of experience in cloud computing, M&M can draw on the services of Azure Machine Learning and the integrated AutomatedML functionality when creating proof of concepts. By using these tools, M&M is able to achieve its goals quickly and efficiently while saving costs. Azure Machine Learning is a fully managed cloud service platform that simplifies the creation, training and deployment of machine learning models. The integrated AutomatedML feature provides automated modeling that replaces manual configuration of models and hyperparameters. This allows M&M to reduce development time and increase efficiency without compromising the quality of the results.

If you would like to find out more about the possible applications and functions of AutoML, please contact our team of experts.

About the author

 

As part of the Data & AI team, Lukas Hauser supports companies in the development of data-driven business models and the implementation of machine learning models. In his bachelor's thesis, he compared and evaluated various AutoML offerings. From October 2023, he will start his part-time Master's degree in Artificial Intelligence to further expand his skills and knowledge in this field. 

Created by