Automated Machine Learning

Automated machine learning studies how to choose and improve the steps that turn data into a useful model, from preprocessing and feature selection to algorithms, hyperparameters, and deployment. Traditional approaches include Bayesian optimization, evolutionary search, neural architecture search, and meta-learning that uses dataset characteristics or prior experiments to guide choices. Increasingly, language-model agents generate and revise executable pipelines, using tool feedback, memory, and validation results to automate broader machine-learning engineering tasks. These systems can reduce manual effort, but reliable progress depends on fair compute budgets, leakage-free evaluation, reproducibility, and evidence that improvements generalize beyond the search data.