AutoML, short for automated machine learning, is the broad effort to automate the many decisions involved in building a machine-learning model so that good results can be obtained with less manual expert effort. A typical supervised-learning project requires choosing how to clean and encode the data, which model family to use, and how to set the model’s hyperparameters; AutoML aims to make some or all of these choices automatically.
The modern academic line of AutoML is well represented by auto-sklearn, introduced in the 2015 NeurIPS paper “Efficient and Robust Automated Machine Learning” by Feurer and colleagues, which framed model and preprocessing selection plus hyperparameter tuning as a single optimization problem solved with Bayesian optimization and ensembling. The field spans several layers: hyperparameter optimization (tuning a fixed model), full-pipeline search (choosing preprocessing and model together), and neural architecture search (designing the network itself). Commercial offerings such as Google Cloud AutoML extended the same idea to image, text, and tabular problems for business users.
AutoML does not remove the need for judgment about data quality, problem framing, and evaluation, and it can be computationally expensive. But it consistently produces strong baselines and lowers the barrier to applying machine learning.
For a business reader, AutoML is the technology behind the promise that teams without deep data-science staff can still build custom, reasonably good models.