Meta-Learning: Learning to Learn Fast

• Define the Meta-Learning Problem
  • A Simple View
  • Training in the Same Way as Testing
  • Learner and Meta-Learner
  • Common Approaches
• Metric-Based
  • Convolutional Siamese Neural Network
  • Matching Networks
    • Simple Embedding
    • Full Context Embeddings
  • Relation Network
  • Prototypical Networks
• Model-Based
  • Memory-Augmented Neural Networks
    • MANN for Meta-Learning
    • Addressing Mechanism for Meta-Learning
  • Meta Networks
    • Fast Weights
    • Model Components
    • Training Process
• Optimization-Based
  • LSTM Meta-Learner
    • Why LSTM?
    • Model Setup
  • MAML
    • First-Order MAML
  • Reptile
    • The Optimization Assumption
    • Reptile vs FOMAML
• Reference

 

https://lilianweng.github.io/lil-log/2018/11/30/meta-learning.html#full-context-embeddings

https://www.zhihu.com/question/264595128/answer/743067364