Automated Machine Learning, often abbreviated as AutoML, is the process of automating the task of machine learning model selection, feature engineering, hyperparameter tuning, iterative modeling, and model assessment. This method significantly lowers the hurdles for non-experts to apply machine learning in their work and often results in superior predictive performance.
Imagine you’re trying to bake the best chocolate cake. There are many variables - like the type of chocolate, the amount of sugar, baking time - affecting how good your cake will be. In traditional baking, you will try different combinations and finally find the best one after numerous trials. But what if you have a magical robot that can try out every possible combination for you and settle on the perfect recipe? Automated Machine Learning is like this magical robot but for solving complex problems with data, not baking cakes.
Automated Machine Learning, or AutoML, strives to automate the end-to-end process of applying machine learning to real-world problems. In a typical machine learning application, practitioners have a broad array of algorithms and hyperparameters they can use to solve their problem, but selecting the right algorithm and features can be time-consuming and requires expertise. AutoML simplifies this process by intelligently exploring many possible options, then selecting the best model.
AutoML is not a tool for achieving artificial general intelligence or automating the creation of new types of models. Instead, its goal is to make machine learning accessible to non-experts and improve the efficiency of experts. It streamlines the machine learning pipeline, which includes data preprocessing, feature extraction, model selection, hyperparameter optimization, and model evaluation.
A key functionality provided by AutoML is hyperparameter tuning, which is the process of adjusting the knobs that control the learning process. This can often lead to substantial improvements in model performance. Other essential features are automatic feature engineering (which identifies complex patterns in raw data) and model selection (which determines the most suitable algorithm given the dataset). Best of all, these processes are largely automated, significantly reducing the time and expertise required to develop production-grade machine learning models.
One relevant branch of AutoML is Neural Architecture Search (NAS), which is the process of automating the design of artificial neural networks. NAS is an approach to designing architectures that are more efficient and robust than those that human designers have been able to create.
Machine Learning, Feature Engineering, Hyperparameter Tuning, Neural Architecture Search, Data Preprocessing, Model Selection, Model Evaluation