Weight Space and the PLA Criterion

  1. Where perceptron weight vector W lives
  2. W space
  3. Extension to multioutput perceptrons
  4. Interpretation of training in weight space


  1. Program MAXNET for 10 input problem. (More on MAXNET in Lippmann, IEEE Signal Processing Magazine, April 1987) ANSWER
  2. Numerically find the minimum of y = (x - 1)^2 and y = x^2 - 1 = (x-1)(x+1) using gradient descent. From which initial points does gradient descent reach each minimum? ANSWER

Back to the CS667 Home Page