Towards automating machine learning
summary
Benchmarks run on Iris and a real insurance dataset with XGBoost on 16‑core machines over many random seeds show automated hyperparameter tuning routinely matches or beats an expert the human stops improving after three manual updates on Iris while optimizers keep improving Sequential model based approaches perform best once they have enough samples where Gaussian process overtakes random search and the tree‑structured Parzen estimator and random forests remain simple and surprisingly effective
AI-written summary · about
chapters (10)
- 0:00Automating ML introduction
- 1:50Hyperparameter tuning in ML
- 3:18Grid and random search limits
- 6:18Sequential model-based optimization
- 9:08Proxy models for tuning
- 14:18Algorithm comparison
- 16:06Benchmarking results
- 21:35Key takeaways
- 23:44Automation beyond tuning
- 26:37Q&A
related talks
27mAutomating XGBoost Hyperparameter Tuning with Agentic AI
26mModel Selection with Python: An Introduction to Hyper Parameter Tuning
41mOptimizing Machine Learning Models
25mHyperparameter Tuning and Visualization in Deep Learning
33mBuild Better AI with First-Party Data and Fine-Tuning
25mHumans helping machines help humans run machines
18mBest Practices for Hyperparameter Optimization
18mAlexandra Johnson, Software Engineer, SigOpt, NYC 2017