6.034 Artificial Intelligence - Recitations, fall 2004 online slides on learning

Next: Parameters That Can Affect Previous: When to Consider Neural

Two Computation Phases

  1. Training phase
  2. Testing / use phase

  • During training, run net on examples and compare net's output ($y$) to desired output ($y^d$)

    Weights are adjusted after each training step using function

    \(w^{new}\) = \(w^{old}\) + ($y^d$ - y)x

    Optionally, the threshold can be adjusted as well using function

    \(t^{new}\) = \(t^{old}\) - ($y^d$ - y)

    Notice that x is the value of the input feature, thus weights are changed ONLY for nodes that are activated (used in the computation)