Model types

Convolutional (CNN)

• detection, classification

• y = F(x)
  • x = 224x224x3 pixels
  • y = 1000 clients
  • w - weights
  • b - bias

Recurring (RNN)

• Memory (LSTM) → speech recognition, music generation

  

• WaveNet

  

  

  

  • Encoder-Decoder

GAN

• generating samples (new photos) = imagination. competition between generator + descriminator. graphics, arts.
  • Auto regressive - PixelCNN
  • DeepDream
  • GPT
  • VQGAN + CLIP
  • Imagen
  • Dall-E 2
  • Stable Diffusion

    

  • https://www.youtube.com/watch?v=Z6rxFNMGdn0

    

    • Automatic architecture composiiton:

  

General Intelligence

• Final goal