nn

Neural Networks

• Components
  • Synaptic weights $w_i$
  • Bias
  • Summing function
  • Activation function
    • Logistic
    • Sigmoid
    • Hyperbolic tangent
• Gradient descent
• Layers: input, hidden, and output
• Deep neural networks
• Advantages
  • Multimodal data
  • Closely approximate any function, given enough neurons and training data
• Disadvantages
  • High cost
  • Large dataset required
  • Local optima
  • Hart to interpret