Update Log
2021-07-24: Initial Upload

Auto-Encoder

  • It is an unsupervised approach for learning a lower-dimensional feature representation from unlabelled training data.
    rXY1p2f
  • The encoder is trained usually with a decoder trying to recover the original data xx. L2 loss between the reconstructed and original input data is calculated to facilitate the training process.
    dhgpw6s
  • The encoder alone can initialise a supervised learning without decoder as a feature extractor.
    fosqnsr

Variational Auto-Encoder

  • Why VAE?
    • Samples are discrete and can not represent the whole distribution 100% accurate. Let’s take the moon as an example.
      ORKyiMO

Reference

  1. ML Lecture 18: Unsupervised Learning - Deep Generative Model (Part II)