The following network designed to classify 2-dimensional continuous variables into two classes. Activation functions are ReLU (in the hidden layer) and a Sign function in the output layer. All weights are either +1 or -1 in this network (blue and arrows denote the weight of +1 and -1 accordingly). All the biases are equal to -1.

Given the current weight and type of activation function, mathematically drive the the decision region generated by this network and draw it in a two-dimensional space. Once you found the decision region, give two input examples from two sides of the region and show that the classifier is able to correctly classify them.

Weight on Red Arrows: -1 Weight on Blue Arrows: +1 Figure 2: 3layer feedforward network