Figure 1.Single- and double-node NN results for classifying 100 6x1 vectors for the examples that only had a single 1 in the 6x1 vector (class=1) or not (class=0).

Single and Double-Node Neural Network Lab

OBJECTIVE: Generate training set of 6x1 vectors that contain 1s and 0s. Classify which ones only contain a single 1 (class=1) or not (class=0). Use a single node neural network and a double node neural network.

PROCEDURE:

  1. Load into your working directory Display.m, Display2.m, gradient.m, gradient2.m, TwoNode.m, gradientnn.m, NNode.m, SingleNode_nonlinear.m.

  2. Type "TwoNode" while in MATLAB. Results will be displayed for the predicted classes.

  3. Explain the differnces between the two-node and single-node results.

  4. Change the input to be a 3x1 vector, and change code so it can run this. Rerun code. explain results compared to the input with 6x1 vectors.

  5. Type NNode.m and compare the results to the single and double node results. Change number of layers and comment on results.

  6. Change code so activation functions are for likihood objective function and compare the result with L2 norm. Currently, code is for L2 objective function. Liklihood gradient replaces the gradient calculation at each layer dg=g(z)(1-g(z)) with a 1.
  7. Change activation function from sigmoid to ReLu. ReLu(z) is 0 if z is less than zero. ReLu(z)=z if z>0. The derivative of ReLu is zero for z<0 and is equal to 1 if z>0. Compare ReLu results with the sigmoid results.