← AI Terminology
AutoML - Automated Machine Learning
AutoML stands for Automated Machine Learning: the use of algorithms to automate the process of selecting, building, and tuning ML models — reducing the need for manual expert intervention.
It encompasses feature engineering, model selection, hyperparameter optimisation, and pipeline assembly.
It encompasses feature engineering, model selection, hyperparameter optimisation, and pipeline assembly.
Why It Matters in AI
Building a good ML model requires decisions at every step — which algorithm, which features, which hyperparameters. AutoML automates these choices using search algorithms (Bayesian optimisation, evolutionary search, random search), democratising ML for teams without deep expertise. At the same time, frontier AutoML systems (Google's NAS-powered architectures) have discovered model designs that human experts would not have found.
Key Points
| Aspect | Description |
|---|---|
| HPO | Hyperparameter Optimisation — Bayesian or random search over learning rate, depth, dropout etc. |
| NAS | Neural Architecture Search — automates the design of the network topology itself |
| Key tools | AutoGluon, H2O AutoML, Google AutoML, AutoKeras, TPOT, Azure AutoML |
| Limitation | Best AutoML still loses to expert humans on novel problem types; can't replace domain insight |
| Pipeline search | Searches over combinations of preprocessing, feature engineering, and model steps |
| Feature selection | Auto-selects the most informative features from raw data |
Simple Analogy
AutoML is to model building what a travel search engine is to trip planning: instead of manually comparing airlines, hotels, and routes, you specify a destination and a budget and the system searches the space of options to find the best combination. You still have to evaluate the result.
Common Usage Examples
AutoGluon.fit(train_data, label='target')— fits multiple models, ensembles best performers- Google AutoML Tables: upload a CSV, specify the target, get a trained model with zero code
optuna.create_study(direction='maximize').optimize(objective, n_trials=100)— HPO with Optuna- NAS: EfficientNet discovered by neural architecture search, outperforms hand-designed CNNs
- Azure ML AutoML: integrated with Azure data pipelines, outputs models with explainability reports
Summary
In short: AutoML lets algorithms find good ML pipelines automatically — making ML accessible to non-experts and uncovering architectures experts wouldn't design by hand.