What is AutoML (Automated Machine Learning)?
"AutoML" is the abbreviation for "Automated Machine Learning". AutoML is a new development in the evolution of the Machine Learning (ML). The aim of AutoML is to provide iterative steps in the ML model development to automate in order to efficiently create an optimised model.
How AutoML solutions work
The degree of difficulty of the automation between the individual steps varies, with the steps of model selection and the Hyperparameter-optimisation offer themselves for automation because of their use case independence. Therefore, automating these two steps is what is commonly understood by AutoML. Ideally, the input includes only a cleaned dataset, an error metric and the maximum time to find the best Model. The output is a ranked list of the tuned models, ordered by the error metric.
AutoML solutions are broadly applied in three main categories: standalone code packages, cloud services and specialised Data Science-platforms.
Benefits and advantages
Overall, the use of AutoML solutions increases the productivity of data scientists and reduces the complexity of ML model development by automating mechanical tasks. Specifically, this means that:
- less time is spent on repetitive tasks such as model selection and tuning hyperparameters - leaving more time to focus on the business problem, collect and pre-process helpful data, and communicate the approach and results to stakeholders.
- a solid basis for a proof of concept is provided, which can be further refined in later phases.
If you are interested in a critical examination of AutoML, please have a look at our further reading Blog article over. Stefan Lautenbacher (Senior Data Scientist) uses a stylised workflow of an ML project to discuss the extent to which AutoML can cover the need for human data scientists.
