Autoencoder

(Information Science Expert) Lecture on Cognitive Robotics

• Also known as Autoencoder
• Encode high-dimensional input X into low-dimensional Features z
• Decode low-dimensional feature z into high-dimensional X’
• The Loss Function of the Neural Network is the difference between x and x’
• 
• By changing the decoder to something different from reproducing the same thing, it expands the possibilities
  • Called Seq2Seq
  • For example:
    • Encoder: Image to low-dimensional
    • Decoder: Low-dimensional to text
    • —> Can generate captions for images#image processing#natural language processing
  • For example:
    • Encoder: Japanese to low-dimensional
    • Decoder: Low-dimensional to English
    • —> Machine Translation#natural language processing #Deep Learning