Connectionism

Kinds of Learning Tasks

• Regularity detection, dimensionality reduction, grouping into unspecified classes (unsupervised learning)
  • Simple competitive learning (see also here)
    
  • Kohonen maps
  • Auto-associative feedforward nets (trained with error-driven learning)
• Associative (content-addressable) memory, pattern completion, categorization (into specified classes)
  • Auto-associative feedforward nets trained with error-driven learning (see also here)
  • Hetero-associative feedforward nets trained with error-driven learning
  • Hetero-associative feedforward nets trained with a reinforcement variant of error-driven learning
  • Symmetric recurrent networks (e.g., Hopfield nets (see also here), Boltzmann machines (see also here)), trained with some version of Hebbian learning
• Binding and unbinding symboloids
  • Convolution networks (see also here)

Representing time

• Time-specific input units
• Time-delay units, sliding windows
• Decaying short-term memory implemented using recurrent connections (e.g., Elman net, Jordan net)
• Periodicity and oscillator networks (see also here)

[L645 Syllabus]