AutoML — Automated Machine Learning

AutoML automates the ML pipeline — from data preprocessing to model deployment.

AutoML Pipeline

AutoML automates:
├─ Feature engineering
├─ Model selection
├─ Hyperparameter tuning
├─ Ensemble construction
└─ Pipeline optimization

Tools:
├─ Auto-sklearn: sklearn-based AutoML
├─ H2O AutoML: Enterprise AutoML
├─ Google AutoML: Cloud-based
├─ FLAML: Fast lightweight AutoML
└─ Optuna: Hyperparameter optimization

Neural Architecture Search (NAS)

NAS searches for optimal neural network architecture:

Search space: Possible architectures
Search strategy: How to explore (random, Bayesian, RL)
Evaluation strategy: How to evaluate (full, weight sharing)

DARTS: Differentiable Architecture Search
├─ Makes architecture differentiable
├─ Gradient-based optimization
└─ Much faster than RL-based NAS

Key Takeaways

1. AutoML automates model selection and tuning
2. Auto-sklearn is the best open-source option
3. NAS finds optimal neural network architectures
4. AutoML is competitive with hand-tuned models
5. H2O for enterprise AutoML
6. Optuna for hyperparameter optimization
7. AutoML democratizes ML — less expertise needed
8. AutoML is a starting point, not the end