Automated machine learning, or AutoML, is a set of methods and processes that automate parts of the machine learning workflow, such as feature construction, model selection and hyperparameter tuning, so that models can be built with less manual effort. It aims to make machine learning usable by non-experts and more efficient for experts.
This page is for business and technology leaders who want to know what AutoML automates, which tools exist, where human expertise is still needed and how it connects to AI agents.
The AutoML.org overview describes automated machine learning as research that provides methods and processes to make machine learning available to non-experts, improve its efficiency and accelerate research in the field.
In a manual workflow, feature engineering and model choice are handled by hand. AutoML handles that work in an automated way.
The overview lists several tasks that AutoML targets.
| Task | What AutoML does | Why it matters |
|---|---|---|
| Data preparation and features | Helps clean data and select or construct features | Reduces repetitive manual work |
| Model selection | Chooses a model family together with its hyperparameters | Removes guesswork when trying many models |
| Hyperparameter optimisation | Automates a slow, error-prone tuning task | Gives new problems a repeatable method; see hyperparameter tuning |
| Neural architecture search | Looks for well-performing deep network designs | Developers do not evaluate architectures by hand |
AutoML.org names AutoML systems including AutoWEKA, Auto-sklearn, Auto-PyTorch, AutoGluon, H2O AutoML, MLBox, TPOT and TransmogrifAI. For hyperparameter optimisation it names SMAC3 and DEHB.
Choose by the language and framework your team already uses, and test the tool on a problem where you know a good answer.
The overview says AutoML lets people without machine learning expertise build models off the shelf, and reduces the time developers spend on tuning and architecture trials. It also supports fairer research comparisons: tuning all algorithms with the same toolkit shows whether a method is genuinely better or just better tuned.
The page frames AutoML as making expert tasks accessible, not as removing the need for expertise. Someone still has to define the problem, check the data and judge whether the result is fit for use.
Agents are built on models, and AutoML shortens the path to a first working model. Teams building agents on ElixirClaw, the AI agents computing platform from XenonStack can use AutoML for the predictive components, while keeping each run, its data and its chosen model on record. Another technique worth knowing in this area is bootstrap aggregation.
To map one decision your agents make, and the data and policy it depends on, book a Context Workshop. See what happens in the workshop.
AutoML provides methods and processes that automate parts of building a machine learning model, such as data preparation, model selection and hyperparameter optimisation, so non-experts can build models more easily.
AutoML.org lists data cleaning and feature work, model selection together with hyperparameters, hyperparameter optimisation and neural architecture search.
No. The overview presents AutoML as making expert tasks accessible and reducing time spent on tuning and trials. Expertise is still needed to frame the problem and judge the result.
AutoML.org names Auto-sklearn, Auto-PyTorch, AutoGluon, H2O AutoML, TPOT, AutoWEKA, MLBox and TransmogrifAI, plus SMAC3 and DEHB for hyperparameter optimisation.