Geoffrey Hinton – How Neural Networks Really Work

  • Supervised Learning
    • With CNNs we’ve got all kinds of pattern matching applications:
      • Image Recognition
      • Text Recognition
      • Speech Recognition
      • Automatic Colorization of Black and White Images
    • With RNNs, we’ve got:
      • Encoders for Image and Text, which convert Image/Text input into thought vectors
      • Decoders for decoding Thought Vectors into Image/Text
    • The application of RNNs are:
      • Machine Translation:
        • Encoder: Speech/Text in Language A -> Thought Vector
        • Decoder: Thought Vector -> Speech/Text in Language B
      • Image Captioning:
        • Encoder: Image -> Thought Vector
        • Decoder: Thought Vector -> Speech/Text in any Language
      • Image Search
        • Encoder: Speech/Text -> Thought Vector
        • Decoder: Thought Vector -> Image
      • Automatic Text Generation (Summation)
        • Encoder: Speech/Text (full length) -> Thought Vector
        • Decoder: Thought Vector -> Speech/Text (shorter)
      • Automatically Adding Sounds To Silent Movies
      • Automatic Handwriting Generation
        • Encoder: Speech/Text (full length) -> Thought Vector
        • Decoder: Thought Vector -> Pen Coordinates
      • Text to Images
        • Encoder: Text -> TV (Thought Vector)
        • Decoder: TV -> Image   ref: https://arxiv.org/pdf/1506.03500.pdf
    • Current State for RNNs is a billion or so learned parameters (weights), which in brain is 100K times more
    • Test for Future AI: Trophy didn’t fit in Bag because it was too big/small.
  • RL
    • Automatic Game Play
    • Autonomous Driving
Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s