元学习，英文为Meta learning，意思是learn to learn/learn how to learn，其目的是使机器学习更加automatic, Frictionless, and never-ending。我理解的“元学习”是将“学习”视为元（研究对象）。
Meta: 变化、变换、自指的
Meta data: 关于数据的数据
Meta analysis: 医学，分析的分析，对独立分析的分析
看的第一篇文献：Vanschoren, J. (2018). Meta-Learning: A Survey. 1–29. http://arxiv.org/abs/1810.03548
The challenge of meta learning：to learn from prior experience in a systematic and data driven way。
1.Collect meta-data. The prior learning tasks and previously learned models.
2.Learn from the meta-data, to extract and transfer knowledge which guides the search for optimal models for new tasks. What way of learning from prior experience is easy?
The catagories of meta learning Learning：
1. purely from model evaluation；
2. Characterize tasks to more explicitly express task similarity and build meta models；
3. Transfer trained model parametres betweent tasks that is inheritly similar.
以上分类Based on the type of data they leverage, from the most general and the most task-specific.
现在有一些问题：
1. How many types of learning? What about their parameters?
Machine learning: CNN, RNN, architecture of network, parameters ofnetwork.
Task definition:automatically tune.
2. How to model learning approaches?
Meta: 变化、变换、自指的
Meta data: 关于数据的数据
Meta analysis: 医学，分析的分析，对独立分析的分析
看的第一篇文献：Vanschoren, J. (2018). Meta-Learning: A Survey. 1–29. http://arxiv.org/abs/1810.03548
The challenge of meta learning：to learn from prior experience in a systematic and data driven way。
1.Collect meta-data. The prior learning tasks and previously learned models.
2.Learn from the meta-data, to extract and transfer knowledge which guides the search for optimal models for new tasks. What way of learning from prior experience is easy?
The catagories of meta learning Learning：
1. purely from model evaluation；
2. Characterize tasks to more explicitly express task similarity and build meta models；
3. Transfer trained model parametres betweent tasks that is inheritly similar.
以上分类Based on the type of data they leverage, from the most general and the most task-specific.
现在有一些问题：
1. How many types of learning? What about their parameters?
Machine learning: CNN, RNN, architecture of network, parameters ofnetwork.
Task definition:automatically tune.
2. How to model learning approaches?
