Perceptron   summation(wixi > T)  if  yes 1
                                    else 0.

To train:
   add an additional input of -1  (with an additional weight)

Repeat over training example:  
 If classifies vector perfectly, do nothing.
 If 1 when should be 0, subtract input vector from weights
 if 0 when should be 1, add input vector to weights.

  If the function is "linearly separable" this will learn the function
    (examples of non-linearly separable functions are XOR and PARITY)...

================================================
Decision tree:
    always ask that question which is expected to leave the least remaining
    uncertainty (diversity)

      recurse on subtrees.