Neural Network Learning

CSC 261 - Artificial Intelligence - Weinman



Answer the following questions. Record your answers in your Reading Journal.
  1. Explain how the last equation on p. 726, the loss gradient for learning in logistic regression, relates to the penultimate form in the derivation of the learning rule for output nodes in neural networks on p. 735, which is
    -2(yk-ak)g(ink)aj.
  2. Consider that the modified error Δk is defined as the error at output unit k times the rate of change in the activation of that unit. In your own words, briefly explain the analogy in meaning for the modified error Δj at hidden unit j, as defined in Equation (18.12) .
  3. Explain an example from your own experience of something akin to overfitting.