Meta Learning
In Machine Learning there are so-called meta-tasks:
- Model Selection
- Parameter Tuning
- Estimating model’s ability to generalize to new data
Meta Learning is a set of Machine Learning techniques for addressing these tasks. The most popular are
- Cross-Validation for estimating the prediction quality of models
- Ensemble Learning for creating stronger models by combining several weaker ones
These techniques generate samples from the data and then train and evaluate models based on these samples
They all have two common steps:
- samples are generated from the input data
- Machine Learning models are trained on these samples
Scalable Meta Learning
See the paper by S. Schelter:
- Schelter, Sebastian, et al. “Efficient Sample Generation for Scalable Meta Learning.” (pdf
Links
- http://en.wikipedia.org/wiki/Meta_learning_(computer_science)
- http://www.scholarpedia.org/article/Metalearning for thorough treatment